Canadian Consumers‟ Functional Food Choices:

Labelling and Reference-Dependent Effects

A Thesis

Submitted to the College of Graduate Studies and Research

in Partial Fulfillment of the Requirements

for the Degree of Doctor of Philosophy

In Department of Bioresource Policy, Business and Economics

University of Saskatchewan

by

Ningning Zou

Copyright Ningning Zou, May 2011. All Rights Reserved.

i

PERMISSION TO USE

In presenting this thesis in partial fulfillment for the requirements for a Postgraduate

Degree from the University of Saskatchewan, I agree that the Libraries of the University of

Saskatchewan may make it freely available for inspection. I further agree that permission for

copying of this thesis in any manner, in whole or in part, for scholarly purpose may be granted

by the professor or professors who supervised my thesis work, or in their absence, the Head of

the Department or the Dean of the College in which this thesis was done. It is understood that

due recognition will be given to me and the University of Saskatchewan in any scholarly use that

may be made of any material in this thesis.

Requests for permission to make use of or to copy the material in this thesis in whole or in

part should be addressed to:

Head of the Department of Bioresource Policy, Business and Economics

University of Saskatchewan

51 Campus Drive

Saskatoon, Saskatchewan

S7N 5A8

ii

ABSTRACT

The growing interest among consumers in the link between diet and health makes

functional food one of the fastest growing sectors in the global food industry, especially

functional dairy products. Understanding consumer choices with respect to functional food is an

important and relatively new research area. Given the credence nature of functional food

attributes, labelling plays a key role in allowing consumers to make informed choices about

foods with enhanced health attributes. In 2007, Canada launched a review of the regulatory

system for health claims on functional foods, which included rules concerning the approval,

labelling and verification of health claims. In 2010 two new health claims related to oat products

and plant sterols were approved by Health Canada. An analysis of how consumers respond to

health claim information is therefore timely.

This thesis focuses on examining the effects of different types of labelling and verification

of health claims on consumers stated preferences for a specific functional food product, Omega-

3 milk. The analysis incorporates reference-dependent effects. This study improves the

knowledge of Canadian consumer understanding of health claims and the impact of health

claims on consumer choice. This research is one of the first studies to simultaneously examine

the effects of different types of health claims (e.g. function claims, risk reduction claims and

disease prevention claims) and other ways of signalling or implying health benefits (e.g.

symbols) on Canadian consumers‘ functional food choices. This study contributes to the

knowledge in this domain by providing a comparative analysis of different types of labelling

strategies. The extant knowledge of labelling effects in the formats of risk reduction claims,

disease prevention claims and symbols or imagery on functional foods is limited. One of the

primary contributions of this study is addressing this gap in the literature.

The theoretical framework of this thesis is based on random utility theory. A stated

preference choice experiment is designed to examine consumers‘ response to Omega-3 milk

under different labelling scenarios. Using data from an online survey of 740 Canadians

conducted in summer 2009, discrete choice models, including Conditional Logit, Random

Parameter Logit and Latent Class models, and Willingness-To-Pay (WTP) values are estimated.

The results suggest that full labelling (function claims, risk reduction claims and disease

prevention claims) is preferred over partial labelling (e.g. the use of a heart symbol to imply a

iii

health claim), but primarily for risk reduction claims. There is no significant difference between

a function claim, such as ―good for your heart‖ and partial labelling in the form of a red heart

symbol. The results also suggest that consumers on average respond positively to verification of

health claims by government and the third party agencies, however, the Latent Class models

reveal considerable heterogeneity in consumer attitudes toward the source of verification. The

influences of key-socio-demographic (e.g. income, education and health status) and attitudinal

factors (e.g. attitude, trust and knowledge) provide further insights into consumer responses in

the choice experiment to identify different consumer segments. Moreover, the results reveal

reference-dependent effects where perceived losses of ingredient or price attributes have a

greater influence on consumer choice than perceived gains.

In terms of industry and public policy implications, this study suggests that food

manufacturers in Canada would benefit from the ability to make more precise health claims. The

implications derived from the Latent Class Models could help the Canadian functional food

industry to identify target consumer segments with different characteristics for the purpose of

developing marketing strategies. Furthermore, the results of this study suggest that Canadian

consumers are receptive to both full labelling and partial labelling. It indicates that public policy

makers need to pay attention to effectively regulating health claims for functional foods so as to

balance the need for credible health claims to facilitate the development of the functional food

sector with the imperative of protecting consumers from misleading health claims. Public policy

makers should also be aware that the verification of health claims plays an important role in

reducing consumers‘ uncertainty and making health claims more credible.

iv

ACKNOWLEDGEMENTS

I would like to deeply express my gratitude and appreciation to Dr. Jill Hobbs, who is

such a supportive and respectable supervisor. Throughout my whole Ph.D. program, she was

always patient, guided me to the right directions, encouraged me to pursue my own research

ideas and provided a great amount of valuable suggestions and guidance. I appreciate all her

contributions of time, advice and funding to make me successfully complete the Ph.D. program.

I would like to extend my appreciation to my advisory committee members, Dr. William

Kerr, Dr. Jing Zhang and Dr. Marjorie Delbaere. I appreciate their expert advice, generous time

spending and valuable comments in different aspects of my research areas. I am also grateful to

Dr. Gale West to be my external examiner and provided insightful suggestions during the

defence.

I also greatly appreciate the Canadian Dairy Commission and Consumer & Market

Demand Network for their generous scholarships to offer me the financial support during the

period of my Ph.D. studies.

I would also like to thank all the faculty, staff and graduate students in our lovely

department for their friendships and collaboration. I would like to be grateful to Dr. Dick

Schoney for his good advice and encouragement not only like a professor but also like a good

friend.

I am very thankful to David Di Zhang for his support, encouragement and understanding. I

want to deeply express my gratitude to my parents and my brother for their unconditional love

and supports. They inspire me to overcome all difficulties and pursue the real goals in my life.

v

TABLE OF CONTENTS

Chapter 1: Introduction and Research Issues .......................................................................................... 1

1.1 Introduction and Background.............................................................................................................. 1

1.2 Researchable Problem and Issues ....................................................................................................... 3

1.2.1 A Comparison of Health Claims Regulation in the United States and Canada ........................... 5

1.2.2 The Issue of Partial Labelling and Full Labelling ...................................................................... 12

1.2.3 Other Research Issues: Verification, Lifestyle, and Reference-Dependent Effects ................... 14

1.3 Research Methodology and Organization of the Thesis ................................................................... 16

Chapter 2: Background: the Functional Food Sector and Consumer Research Methodologies ....... 18

2.1 Definitions of Functional Foods ....................................................................................................... 18

2.2 Developments in Functional Food Markets ...................................................................................... 20

2.3 Functional Foods and Information Asymmetry ................................................................................ 23

2.4 Canadian Consumer Attitudes towards Functional Foods: Previous Studies ................................... 27

2.5 Review of Consumer Research Methodologies ................................................................................ 29

2.5.1 Stated Preference Methods versus Revealed Preference Methods ............................................. 29

2.5.2 Contingent Valuation and Conjoint Analysis ............................................................................ 31

2.5.3 Application of Choice Experiments in the Food Sector ............................................................. 32

Chapter 3: Research Methodology: the Choice Experiment ................................................................ 37

3.1 The Selection of Attributes and Levels in the Choice Experiment ................................................... 38

3.2 Choice Experiment Design ............................................................................................................... 42

3.3. The treatment of the ‗No-purchase‘ Option ..................................................................................... 50

3.4 Survey and Data Collection .............................................................................................................. 51

Chapter 4: Econometric Models and Descriptive Analysis ................................................................... 56

4.1. Econometric Models and Estimation Methods ................................................................................ 56

4.2 Descriptive Analysis ......................................................................................................................... 69

vi

4.2.1 Sample Characteristics ............................................................................................................... 69

4.2.2 Attitudinal Characteristics .......................................................................................................... 72

4.3 Factor Analysis ................................................................................................................................. 78

Chapter 5: Results: the Effects of Labelling on Functional Food Choices .......................................... 83

5.1 Base Model: CL Estimates and WTP ............................................................................................... 86

5.2 CL Model with Interaction Effects ................................................................................................... 91

5.3 CL Model with Interaction Effects for Covariates and Key Factors................................................. 94

5.4 Base Model: the Random Parameter Logit Model and WTP ......................................................... 105

5.5 Base Model: LCM Estimates .......................................................................................................... 109

5.6 LCM Model with Interaction Effects of Covariates and Key Factors ............................................ 114

5.7 The LCM Model with Class Membership Indicators ..................................................................... 120

5.8 Conclusion ...................................................................................................................................... 125

Chapter 6: Reference-Dependent Effects in Consumers‟ Functional Food Choices ........................ 130

6.1 Introduction ..................................................................................................................................... 131

6.2 Prospect Theory .............................................................................................................................. 136

6.3 The Reference-Dependent Model ................................................................................................... 137

6.3.1 Modelling Reference-Dependent Effects ................................................................................. 140

6.4 Data and Econometric Model ......................................................................................................... 146

6.5 Estimation Results .......................................................................................................................... 151

6.6 Conclusions ..................................................................................................................................... 160

Chapter 7: Implications and Conclusions ............................................................................................. 164

7.1 Summary of Major Research Findings ........................................................................................... 165

7.1.1 Full Labelling and Partial Labelling ........................................................................................ 165

7.1.2 Effects of Different Verification Organizations ....................................................................... 167

7.1.3 The Role of Attitudinal and Demographic Information ........................................................... 168

7.1.4 Reference-Dependent Effects................................................................................................... 169

vii

7.1.5 Choosing a Proper Discrete Choice Model .............................................................................. 171

7.2 Implications..................................................................................................................................... 172

7.2.1 Implications for the Functional Food Industry ......................................................................... 172

7.2.2 Public Policy Implications ....................................................................................................... 175

7.3 Limitations and Future Research .................................................................................................... 178

References ................................................................................................................................................ 181

Appendix 1: Post Hoc Estimation for the CL Model with Interaction Effects .................................. 194

Appendix 2: Estimation Results of the RPL Model Including Reference-Dependent Effects ......... 195

Appendix 3: Post Hoc Estimation for the Unused/Unqualified Data Set ........................................... 197

Appendix 4: The Survey ......................................................................................................................... 205

viii

LIST OF TABLES

Page

Table 1.1 An Analytical Framework for Functional Food Health Claims in

Canada

9

Table 3.1 Attributes and Levels in Choice Experiment 40

Table 3.2 An Example of a Choice Set from the Survey 49

Table 4.1 Socio-Demographic Characteristics of the Survey Participants and

the Canadian Population

70

Table 4.2 The Key/Component Factors in the Factor Analysis 81

Table 5.1 Summary of the Variables for the Estimation Models 84

Table 5.2 Base Model: CL Estimations and WTP 87

Table 5.3 Wald Test for Difference in WTP 89

Table 5.4 CL Model with Interaction Effects between the Main Attributes 92

Table 5.5 (a) CL Model with the Covariates and Key Factors 95

Table 5.5 (b) WTP of CL Model with the Covariates and Key Factors 96

Table 5.6 Base Model: RPL and WTP Estimates 106

Table 5.7 Respondents‘ Heterogeneous Preferences for the Attributes with

Random Parameters

108

Table 5.8 Base Model: LCM Estimation Results 110

Table 5.9 (a) LCM with Interaction Effects for the Covariates and Key Factors 115

Table 5.9 (b) WTP of LCM with Interaction Effects for the Covariates and Key

Factors

116

Table 5.10 LCM with the Class Membership Indicators 122

Table 6.1 Variable Descriptions for the Reference-Dependent Model 147

Table 6.2 (a) Conditional Logit Estimates with Reference-Dependent Effects 153

Table 6.2 (b) WTP Estimates with Reference-Dependent Effects 154

ix

LIST OF FIGURES

Page

Figure 1.1 Partial labelling examples 14

Figure 2.1 Number of Canadian Companies with Functional and Health

Products in the Market

23

Figure 4.1 A Comparison of Socio-Demographic Distributions between the

Sample and the Canadian Population

71

Figure 4.2 Health Information Sources and Trust in those Information Sources

73

Figure 4.3 Top Ranked Health Information Sources 75

Figure 4.4 The Frequency of Canadian Consumers‘ Functional Food

Consumption

77

Figure 6.1 An Illustration of a Value Function 138

Figure 6.2 Reference-Dependent Effects for Price and Omega-3 157

1

Chapter 1: Introduction and Research Issues

1.1 Introduction and Background

Consumers are increasingly interested in the health benefits of foods beyond basic nutrition,

and are giving more appreciation to the potential disease risk reduction and health improving

properties of functional food products (Agriculture and Agri-Food Canada, 2009). There is

increasing awareness of the link between health and diet (Anders and Moser, 2010; Malla,

Hobbs and Perger, 2007; Cash, Goddard and Lerohl, 2006; West et al., 2002), while the

economic and social costs of diet-related diseases, such as cancer, cardiovascular disease and

diabetes, have been growing at a rapid pace (World Health Organization, 2002 and 2005). The

American Dietetic Association (Bloch and Thomson, 1995) points out that accumulating

scientific evidence supports the role of functional foods in the prevention and treatment of at

least four types of leading diseases: cancer, diabetes, cardiovascular disease and hypertension. It

has been claimed that functional foods provide an opportunity to improve the health of

Canadians, reduce health care costs and support economic development in rural communities

(Agriculture and Agri-Food Canada, 2009).

Given the broad range of functional food attributes, there is no simple, commonly

accepted and exclusive definition for the term ‗functional food‘. Different countries use

different definitions in their regulatory systems. Researchers appear to have a variety of

understandings of the concept, but they all recognize ‗health benefits‘ as the main contribution

of functional foods. For example, calcium enriched milk products are widely recognized as able

to help maintain healthy bones and reduce the risk of osteoporosis (Canadian Food Inspection

Agency, 2009). From the consumers‘ point of view, the potential health benefits of functional

2

foods could affect their consumption decisions: an individual consumer might over consume

unhealthy foods or under consume healthy foods. In making a decision regarding the

consumption of functional food, consumers face two types of uncertainty: uncertainty about the

health attributes of a specific food and uncertainty over future health outcomes. Given the

information asymmetry and credence nature inherent in functional foods, labelling (e.g. health

claims) plays a key role in allowing consumers to make informed choices (Roe, Levy and Derby,

1999; Hailu, et al, 2009; Garretson and Burton, 2000; Wansink, 2003; Kozup, Creyer and

Burton, 2003).

In 2007, Canada launched a review of the regulatory system for health claims on functional

food products (Health Canada, 2007). Health Canada planned to update its' assessment

framework to make the system clearer and the claims more credible. In a news article, Crowley

(2008) comments that: ―the current assessment process in Canada has been criticized as being

slow, unpredictable, and restrictive as well as unclear concerning what types of health claims

need to be submitted for consideration and the process to be used to evaluate them‖. Regulation

of functional foods includes rules concerning approval, labelling and verification of health

claims (Health Canada, 1998a). Scientific support for health claims is also important to provide

consumers with products that they can trust (Basu, Thomas and Acharya, 2007).

The challenges of changing the regulatory system for health claims are twofold. First, if the

process for approving claims is relaxed, claims may not be credible and it is hard to protect

consumers against harmful products and/or fraudulent claims. On the other hand, if the

regulatory environment is too restrictive, it might restrict consumers‘ access to functional foods

with health benefits and limit the development of this sector of the food industry. The benefits

associated with revising policies for health claims include potentially improving the diet of

3

individual consumers by facilitating access to foods with proven health benefits; second, creating

a more effective and credible food label system; and third, strengthening the country‘s

competitiveness in the rapidly growing global functional food market.

1.2 Researchable Problem and Issues

There has been a dramatic expansion in the functional food sector of the global food

market over the past decade. It has been claimed that access to functional foods, as well as

adequate labelling of the health effects of functional foods are necessary keys to the growth of

the Canadian functional food market (Basu, Thomas and Acharya, 2007). Regulation of health

claims on food products needs to balance the twin goals of protecting consumers from

misleading or unsubstantiated claims, while providing sufficient information to help consumers

make informed decisions. Some commentators have argued that the Canadian health claim

recognition system is more restrictive than that in other countries such as the United States,

Japan and the European Union (L‘Abbe et al., 2008 and Crowley, 2008). For example, currently,

Health Canada allows seven science-based risk reduction claims (Health Canada, 2010a) to be

used on food labels or in advertisements, compared with 17 health claims (FDA, 2009b and

2009c) in the United States. Among those seven acceptable science-based risk reduction claims

in Canada, two of them were recently approved in 2010. The two newly approved claims

suggest that oat products and plant sterols may reduce health risks by lowering blood cholesterol.

It seems likely that additional health claims could be approved in Canada in the future, and

therefore an analysis of how consumers respond to health claim information is timely. To update

the policies for health claims on foods, both Canada and the United States have undertaken

public consultations with consumers, the food industry and health professionals.

4

This study focuses on examining the effects of labelling on Canadian consumers‘

functional food choices by examining the effects on consumer choices of different types of

health claim or ways of signalling health benefits. Using the data collected from a national

survey, this study explores how changes in labelling formatting (referred to in this study as ‗full

labelling‘ and ‗partial labelling‘) could alter individuals‘ preferences. In this study, formally

stated health claims in an explicit way (e.g. function claims, risk reduction claims and disease

prevention claims) are hereafter referred to as ―Full Labelling‖; and the use of an image or a

visual cue to implicitly imply a health benefit is hereafter referred to as ―Partial Labelling‖.

Furthermore, this study also examines the existence of reference-dependent effects in Canadian

consumers‘ functional food choices, a concept developed from psychology. Reference-dependent

effects are used to measure the value of changes involving a gain or a loss compared with a

reference point and changes to the reference point could lead to reversals of preference (Tversky

and Kahneman, 1991).

In the literature, health claims are widely discussed for regulation issues (Smith,

Marcotte and Harrison, 1996, Fitzpatrick, 2004 and L‘Abbe et al., 2008). In the consumer

analysis area, function claim and risk reduction claim are often examined in the study (Roe et al.,

1999; Kleef, 2005; Hovde et al., 2007; and Bond, Thilmany and Bond, 2008). Very few studies

have examined consumer acceptance of disease prevention claim, a more restrictive claim, and

symbols, the major labelling format for functional food in Canada. This study aims to fill this

gap. This research is one of the first studies to examine simultaneously the effects of different

types of health claims (e.g. function claims, risk reduction claims and disease prevention claims)

and other ways of signalling health benefits (e.g. symbols) on Canadian consumers‘ functional

food choices. The extant knowledge of the labelling effects in the formats of risk reduction

5

claims, disease prevention claims and symbols on functional foods is limited in the literature.

This research aims to fill in this gap. This study improves the knowledge of Canadian consumer

understanding of health claims and the impact of health claims on consumer choice. More

specifically, the research questions explored in this thesis include: (1) how do Canadian

consumers respond to full labelling and partial labelling as signals of a health benefit, (2) how do

they respond to the verification of health claims by different organizations, (3) how do other

attitudinal, behavioural and socio-demographic factors affect consumers‘ choice behaviours, and

(4) whether reference-dependent effects exist when consumers make functional food

consumption choices. Functional food (Omega-3 milk) and regular food (conventional milk) are

used as examples to explore these research questions.

The next section begins by observing that the regulatory environment in Canada is different

from the case in the United States in terms of the number of allowable health claims and the

extent to which the use of health claims is subject to regulatory oversight. Section 1.2.2 explores

the concepts of partial versus full labelling, while section 1.2.3 expands on the other researchable

issues around the source of verification and the effect of behavioural and attitudinal factors.

Section 1.3 briefly outlines the research methodology. The chapter concludes with an overview

of the structure of the thesis.

1.2.1 A Comparison of Health Claims Regulation in the United States and Canada

The United States currently permits two types of claims on functional food products,

function claims and risk reduction claims. First, a product may carry a function claim linking a

substance to an effect on the functioning of the body, called a ‗structure/function claim‘, for

example, ―calcium builds strong bones‖. This type of health claim must be truthful and not

6

misleading and are not pre-reviewed or authorized by the Food and Drug Administration (FDA)

(FDA, 2009a). Second, a health claim which links a nutrient to a particular disease risk reduction,

called a ‗risk reduction claim‘ is permitted. It must be reviewed and evaluated by FDA prior to

use on food labels. Note that these types of health claims are limited to claims about disease risk

reduction, and cannot be claims about the diagnosis, cure, mitigation, or treatment of disease (e.g.

disease prevention claims) (FDA, 2009a). The FDA approach to regulating risk reduction claims

are based on the strength of scientific evidence. There are two types of risk reduction claims in

the United States, unqualified and qualified health claims (FDA, 2009a). The FDA of the United

States has consulted the public on the effectiveness of their qualified and unqualified health

claims since 2002.

The FDA evaluated the scientific evidence supporting risk reduction claims in food and

dietary supplement labelling. The FDA is currently adopting a four level system. FDA level A

health claims are unqualified, reflecting ―significant scientific agreement‖ (―SSA‖) about the

validity of the disease-diet relationship, and levels B, C, D are qualified health claims where the

evidence is progressively weaker (Federal Trade Commission, 2006). For example, ―Diets rich in

calcium may reduce the risk of osteoporosis.‖ (Level A, unqualified health claims); and ―Omega-

3 fatty acids may reduce the risk of heart disease but the scientific evidence is promising but not

conclusive.‖ (Level B, qualified health claims). Currently through the FDA‘s health claims

regulation system, 17 types of unqualified health claims (FDA, 2009b) are allowed for functional

food products in the United States. Roe, Levy and Derby (1999) found that the presence of health

claims on food labels could increase consumers‘ purchase intentions and made them consider the

product healthier. Some foods and supplements are sold with FDA pre-approved health claims,

and functional foods are often sold on the basis of structure/function claims in order to avoid

7

FDA pre-approval requirements that apply to health claims (Heller, Taniguchi and Lobstein,

1999).

The current Canadian regulations on functional food health claims are evolving. In 1995,

two earlier studies commissioned by Agricultural and Agri-Food Canada (AAFC, 1995) and

Foreign Affairs and International Trade Canada (FAIT, 1995) addressed Canada‘s regulatory

issues related to health claims on functional food. In 1996, another research commissioned by

Agricultural and Agri-Food Canada further conducted a comparative analysis of the regulatory

framework affecting functional food development and commercialization in Canada, Japan, the

European Union and the United States (Smith, Marcotte and Harrison, 1996). In recent years, the

number of approved science-based risk reduction claims for use on functional food labels or in

advertisements in Canada increased from five to seven in 2010. Unlike the United States, the

Canadian government not only regulates the assessment and approval of disease risk reduction

claims, but also sets conditions around the use of function claims. Minimum levels and content

requirements need to be reviewed before a function claim is deemed acceptable and allowed to

be used on a product label (Canadian Food Inspection Agency, 2009). Health Canada currently

allows seven specific science-based disease risk reduction claims that reflect the following

relationships (Health Canada, 2003; Canadian Food Inspection Agency, 2009; Health Canada,

2010a):

1. ―a diet high in potassium and low in sodium, and the reduced risk of hypertension;

2. a diet adequate in calcium and vitamin D, and the reduced risk of osteoporosis;

3. a diet low in saturated fat and trans fat, and the reduced risk of heart disease;

4. a diet rich in vegetables and fruit, and the reduced risk of some types of cancer; and

5. maximal fermentable carbohydrates in gum, hard candy or breath-freshening products, and

the reduced risk of dental caries.‖ (Canadian Food Inspection Agency, 2009, p8-7)

6. plant sterols and the lowing of blood cholesterol (Health Canada, 2010c);

7. oat products and the lowing of blood cholesterol (Health Canada, 2010b).

8

The above seven statements are based on the existing unqualified health claims approved by

the FDA in the United States (FDA, 2009b). Health Canada reviewed those claims and

determined that seven of them would be allowed in Canada, while the others remain

unapproved1. Note that the last two risk reduction claims were approved by Health Canada in

2010. These two claims are related to plant sterols and oat products.

Generally speaking, there are three steps for a disease-related health claim on functional

food to be approved by Health Canada (Health Canada, 2010b). First, an application needs to be

submitted to Health Canada to request an approval for the use of certain disease-related health

claims. Taking the oat products for example, in 2007 Health Canada‘s Food Directorate received

a submission from industry requesting the approval for the use of a disease risk reduction claim

on oat products. Second, Health Canada needs to satisfy itself that there is sufficient scientific

evidence to support the claim. The petitioner for the oat products provided Health Canada the

1995 health claim petition which was submitted to the FDA of the United States on oats and

coronary heart disease (which claim was subsequently approved by the FDA in 1997) and other

scientific evidence. Third, after reviewing the evidence, Health Canada makes a final decision.

In November 2010 Health Canada concluded that the following statements may be used on food

labels or advertisements of oat products that meet the qualifying criteria: "[serving size from

Nutrition Facts table in metric and common household measures] of (Brand name) [name of

food] [with name of eligible fibre source] supplies/provides [X % of the daily amount] of the

fibres shown to help reduce/lower cholesterol" (Health Canada, 2010b). Or a short statement in

1 The unapproved unqualified health claims include: ―dietary fat and cancer; folate and neural tube defects; fiber-

containing grain products, fruits, and vegetables and cancer; fruits, vegetables, and grain products that contain fiber, particularly

soluble fiber, and risk of coronary artery disease; dietary saturated fat and cholesterol, and risk of Coronary Heart Disease; soy

protein and risk of Coronary Heart Disease; whole grain foods and risk of heart disease and certain cancers; potassium and the

risk of high blood pressure and stroke; fluoridated water and reduced risk of dental caries‖ (FDA. 2009b).

9

addition to the primary statement can be used, such as: ―Oat fibre helps reduce/lower cholesterol,

(which is) a risk factor for heart disease‖.

Table 1.1 An Analytical Framework for Functional Foods Health Claims in Canada

Structure/

Function

Claimsa

Risk Reduction Claims Therapeutic Claimsa,b

(Disease Prevention

Claims) Product

Specific Generic

DEFINITION Asserts the role of

a nutrient or other

dietary component

intended to affect

a specific structure

or physiological

function in

humans.

Asserts a

relationship

between a

specific food

product and a

reduced risk of a

disease or

condition.

Asserts a

relationship between

a nutrient or other

dietary component in

a diet, and a reduced

risk of a disease or

condition.

Asserts a relationship

between a nutrient or

other dietary component

and the cure, treatment,

mitigation or prevention

of a disease, disorder or

abnormal physical state.

TARGET

GROUP General

population and

sub-groups

General

population and

sub-groups

General population

and sub-groups Specific individuals with

a disease (cure/treat/

mitigate) General

populations and sub-

groups (prevention)

Regulatory

Mechanism Exemption by

Regulation for

Food

Exemption by

Regulation for

Food

Exemption by

Regulation for Food Food and Drug

Regulations - as at

present

Standards of

Evidence c

See Guide to Food

Labelling and

Advertising-

Health Claims

See Guide to

Food

Labelling and

Advertising-

Health Claims

See Guide to Food

Labelling and

Advertising-Health

Claims

As outlined in Food and

Drug Regulations

a Generic and product-specific claims may be applicable here.

b Includes prevention claims.

c modified

by Chapter 8-Health Claims, Guide to Food Labelling and Advertising, Canadian Food Inspection

Agency

Source: Adapted from Health Canada (1998a) and Canadian Food Inspection Agency (2009)

Table 1.1 describes the regulatory framework governing health claims for functional foods

in Canada (Health Canada, 1998a). Developed by Health Canada in 1998 and updated by the

Canadian Food Inspection Agency in 2009, this framework describes the Canadian regulatory

10

environment with respect to the definition, target group, regulatory mechanism and standards of

evidence, for the three types of health claims which are functional claims, risk reduction claims

and therapeutic claims2 /disease prevention claims. Products claiming to treat, cure, mitigate or

prevent a disease or illness are regulated as drugs. Disease prevention claims, linking

consumption of a food to the prevention of a specific disease, are not permitted on food products

either in Canada or the United States, although some scientific studies have shown supportive

evidence (not conclusive) for the role of functional food in preventing certain chronic diseases

(e.g. Bloch and Thomson, 1995).

In term of the regulatory mechanism, function claims and risk reduction claims are

―exemption by regulation for food‖, which means these two types of health claims are exempted

by Food and Drug Regulations and can be used on regular food products, while disease

prevention claims are not exempted from drug claims and cannot be used directly on food in

Canada. With respect to standards of evidence, if a product is claimed to have a health benefit, it

must be supported by scientific evidence to demonstrate its efficacy. According to the risk-

management framework of Health Canada (1998a), the standard of evidence must be

proportionate to the degree of risk. For example, if a food product claims to prevent cancer

(disease prevention claim), that claim should require the highest standards of evidence. However,

if a food product is claimed to be good for the heart (structure/function claim), then a lower

standard of evidence is sufficient. The Canadian Food Inspection Agency (2009) has updated the

standards of evidence related to function claims and risk reduction claims for functional foods

selling in the Canadian market. One aspect of this thesis examines Canadian consumers‘

2 ―Therapeutic claims are claims about treatment or mitigation of a health-related disease or condition, or about restoring,

correcting or modifying body functions. At present, no therapeutic claims have been approved for food in Canada.‖ (Canadian

Food Inspection Agency, 2009, p8-1)

11

response to the three types of health claims, which could inform future developments in the

regulatory system with respect to the effect of different types of information on consumers‘

decisions.

The product examined in this research is Omega-3 milk (see details in chapter three). Of

note is that the risk reduction claims for two types of Omega-3 fatty acids are already allowed by

the FDA in the United States, but not by Health Canada. In 2004, the FDA announced qualified

health claims for omega-3 fatty acids. The following health claims are allowed on food labels or

in advertisements in the United States, “Supportive but not conclusive research shows that

consumption of EPA and DHA Omega-3 fatty acids may reduce the risk of coronary heart

disease‖ (FDA, 2004).

Due to the differences in government regulations, a number of products carrying health

claims in other countries, such as in the US, cannot be labelled with the same health claims when

sold in Canada. Canadian consumers may purchase such food products but they cannot be

marketed in Canada as having the functional attribute. Clearly, the regulatory system needs to

balance consumer protection and the avoidance of spurious or misleading health claims, with the

ability for food manufacturers to communicate the genuine health benefits of functional foods to

consumers. At the present time, the range of products identified as functional foods is therefore

limited. This research aims to inform policy decisions governing health claims to better

understand Canadian consumers‘ reactions to different types of health claims on functional food

products and the effect of verification by different parties.

12

1.2.2 The Issue of Partial Labelling and Full Labelling

Understanding consumer choices with respect to functional foods is an important new area

of research. With the growing scientific evidence regarding the effectiveness of functional

ingredients in enhancing health, reducing the risk of diseases and perhaps preventing some

chronic diseases, it is timely to consider the effects of different types of labelling for functional

foods. The interesting research questions lie in understanding the labelling context: how do

consumers make decisions in environments of uncertainty; which sources of information are

credible; how do different information environments affect willingness to pay; and how do

different perceived health effects affect consumer behaviour with respect to functional foods.

In the absence of labelling, the health benefits for functional foods are a credence attribute,

since the differences between a functional food and a regular food cannot be detected by search

or experience behaviours. Given the importance of labelling and information asymmetry in

understanding the consumer decision-making process for functional foods, the effect of different

labelling strategies should be examined; in particular, whether different labelling policies could

offer sufficient information to affect consumers‘ consumption decisions.

‗Full labelling‘ in this study means direct and explicit health claims on product labels (e.g.

―calcium may reduce the risk of osteoporosis‖), which includes function claims, risk reduction

claims and disease prevention claims, as described in Table 1.1. For example, for Omega-3

functional foods products, none of the three types of full health claims is allowed on food labels

in Canada. However, both function claims and risk reduction claims for Omega-3 fatty acids

have been allowed on food labels in the United States since 2004. Although both partial

labelling and full labelling can be used to carry health information for functional foods, it is

13

possible that full labelling could communicate the information more directly and accurately than

partial labelling. Therefore, this thesis seeks to examine consumer responses to the use of full

labelling by three types of health claims on Omega-3 milk products.

The relatively restrictive regulatory environment for labelling in Canada has resulted in

food companies using an image or visual cue to implicitly signal a health benefit, although some

rules and concerns still apply to use symbols (e.g. heart symbol) on food labels or in

advertisement in Canada (Canadian Food Inspection Agency, 2009). This thesis uses the term

‗partial labelling‘ to describe this strategy. Partial label means the use of a symbol or a visual

cue on the products to implicitly imply health information to consumers as compared with ‗full

labelling‘ which means the health claims are explicitly stated on food labels, such as function

claims, risk reduction claims and disease prevention claims.

Kozup, Creyer and Burton (2003) found that when a heart-healthy symbol was present,

consumers generally believed that the food would reduce the likelihood of heart disease or

stroke. Currently, ‗partial labelling‘ has become a significant labelling format in the Canadian

functional food market. Figure 1.1 provides examples: (1) the use of a red heart symbol on a

soybean milk package to imply that consumption of the product is good for heart health; (2) the

picture of a stomach with an arrow featured on a probiotic yogurt by ACTIVA a Danone brand

to imply that consumption of the yogurt can help digestive health. Informal consultation with a

group of regular yogurt consumers suggested that this imagery was rather obtuse. If more

explicit health claims for probiotic products were allowed in the Canadian food market,

presumably food companies would use a direct health claim with a clear meaning rather than

using an image that may be misinterpreted by consumers. On the other hand, if symbols (partial

labelling) can convey health benefits effectively, such as the red heart symbol on the soybean

14

milk in figure 1.1, those symbols may in fact help consumers to better understand the health

benefits of the products. It is possible that partial labelling could convey the same amount of

information, or even work better than full labelling to inform consumers about the health

benefits of the products (perhaps because of the attractive design or if images are easier to

understand than a written health claim which consumers may not take the time to read). Of

course, it is also possible that ‗partial labelling‘ is used to imply a health benefit for which there

is scant scientific evidence. Therefore, this thesis seeks to explore consumer responses to the use

of partial labelling vs. full labelling within the context of the Canadian regulation environment

governing health claims on food products.

Figure 1.1: Partial labelling examples

1.2.3 Other Research Issues: Verification, Lifestyle, and Reference-Dependent Effects

Verification of health claims is another important issue, which could affect consumers‘

perceived qualify for credence attributes. Given the fact that the health benefits from functional

foods are credence attributes, verification of health claims can help to reduce consumer

uncertainty and make health claims more credible. Only a few studies have been done for this

topic in the functional food sector in Canada, especially at national level (Caswell, Noelke and

15

Mojduszka, 2002; Hailu et al., 2009; and Innes and Hobbs, 2011). The effect of verification by

different organizations is therefore examined in this study. A government organization (e.g.

Health Canada) and a third party organization (e.g. the Heart and Stroke foundation) are

selected as the verifying organizations examined in this study. For example, food products may

contain symbols or images implying an endorsement or healthy diet recommendation by a third

party, such as the Canadian Diabetic Association, or the ‗Health Check‘ program of the Heart

and Stroke Foundation. While government verification is likely to be backed by scientific

evidence, this may not be the case for a third party such as the Heart and Stroke Foundation

(H.S.F.). Nevertheless, these organizations may be still credible. Understanding how consumers

react to public sector (government) vs. third party (H.S.F.) verification is a topic explored in this

thesis.

This study also evaluates the effects of consumers‘ attitudes and lifestyle behaviours on

their functional food choices. Consumers‘ characteristics, attitudes towards functional foods and

diet and exercise behaviours have been recognized in the literature as factors which could affect

their healthy food choices (West et al., 2002, Lusk and Parker, 2009; Hailu et al., 2009; Verbeke,

2005; Cox, Koster and Russell, 2004; DeJong et al., 2003, and Gilbert, 2000). This study

explores how demographic characteristics (e.g. income, education, gender and heart disease) and

attitudinal and behavioural factors (e.g. attitudes towards functional foods, trust in health claims

and nutrition labels, and health knowledge) help account for the heterogeneity of Canadian

consumers‘ preferences for Omega-3 milk.

In addition, the area of reference-dependent effect might be also helpful to further explain

consumer‘s choice behaviour (Tversky and Kahneman, 1991; Hardie, Johnson and Fader, 1993

and Hu, Adamowicz and Veeman, 2006), while, the application of this topic in the domain of

16

functional food is very limited. The Reference-Dependent model was originally developed from

psychology (Prospect Theory) as an approach to explain consumers‘ preferences. The central

assumption of the theory is that losses and disadvantages have greater impact on preferences

than gains and advantages (Tversky and Kahneman, 1991). In Prospect Theory, outcomes are

expressed as changes (gains or losses) from a neutral reference point. There are two important

properties of this theory: reference dependence and loss aversion. Reference dependence means

individuals‘ preferences depend on a reference point; a change in a reference point might leads

to reversals of preference. Loss aversion means losses dominate gains, and people will work

harder to avoid losses than to obtain gains.

The reference-dependent approach informs our understanding of how consumers make

choices about food products with health claims. Reference-dependent effects reflect the fact that

individuals with different reference points could make heterogeneous choice decisions. Thus, a

utility function need to be constructed relative to the reference condition. Chapter 6 takes random

utility theory as the underlying theoretical framework and uses stated preference survey data to

examine the reference-dependent and loss aversion effects in Canadian consumers‘ functional

food choices. It focuses on testing the reference-dependent effects of the functional ingredient

(e.g. Omega-3) and price attributes. The sources of heterogeneity for reference-dependent effects

are also examined through several demographic and attitudinal variables.

1.3 Research Methodology and Organization of the Thesis

The theoretical framework of this thesis is based on random utility theory. A stated

preference choice experiment is used to examine consumers‘ response to Omega-3 milk under

different labelling scenarios. Chapter 2 provides an overview of the functional food market and

17

reviews the literature examining consumers‘ functional food choices. Chapter 3 describes the

design of this consumer survey and choice experiment. Chapter 4 presents the econometric

models and descriptive analysis of the survey data. The analysis uses Discrete Choice models,

incorporating Factor Analysis. Chapter 5 presents empirical estimation results of the effects of

labelling on Canadian consumers‘ functional food choices. Chapter 6 presents the Reference

Dependent model and Prospect Theory and examines reference-dependent effects in consumers‘

functional food choices. Chapter 7 presents conclusions, discusses the limitations of the study

and provides suggestions for future research.

18

Chapter 2: Background: the Functional Food Sector

and Consumer Research Methodologies

This chapter focuses on introducing different definitions of functional foods, describing

the development of functional food markets, and reviewing consumer research literature with

respect to information asymmetry, consumer preference methodologies and the application of

those methodologies to functional food products.

2.1 Definitions of Functional Foods

When Canadian consumers are asked, ―have you ever heard about functional foods?‖ we

can expect a range of responses from puzzlement, to vague familiarity, to knowledgeable

interest. While relatively few consumers are likely to have a clear idea about the term

‗functional foods‘, a substantial number of them have probably consumed functional foods at

some point in the recent past, whether knowingly or not. Increasingly, a wide range of

functional food products are available in Canada, such as orange juice fortified with calcium,

probiotic enriched yogurt, omega-3 fatty acid enriched milk, vitamin C-enhanced soft drinks,

and breakfast cereals enriched with minerals, etc. Functional foods have become recognized as a

separate food category only in recent years in Canada and the United States, while they have

been popular in Europe and Japan since the 1990s. The central characteristic of functional foods

is the pro-active ‗health benefits‘ that the food imparts; placing the food category between

normal food and medicine.

What are functional foods? The central idea is that they are foods with health benefits.

There are numerous definitions of functional foods in the literature. Health Canada (1998b)

defines functional foods as a food product that is consumed as part of a usual diet, and has

19

demonstrated physiological benefits and/or reduces the risk of a chronic disease beyond a basic

nutritional function. Functional food includes both traditional food products and food

innovations derived through conventional breeding methods or through genetic engineering.

Poulsen (1999) categorizes four ‗enrichment‘ methods by which functional foods can be

produced: upgrading, by adding more of a substance that the food already contains; substitution,

by replacing a substance with a similar but healthier component; enrichment, by adding an

ingredient not normally found in the food product; and elimination, by removal of an unhealthy

ingredient from a food.

Doyon and Labrecque (2008) summarized 26 different definitions in the literature and

identified a broadly accepted definition of functional foods. The authors found that four general

concepts should be included in a definition: health benefits, the nature of the food, function and

regular consumption. First, the concept of health benefits is the central component as illustrated

by the 26 selected definitions. There were three specified health benefits in the definitions,

which are physiological benefits (4/26), reducing the risk of diseases (7/26) and preventing

health problems (3). Second, the nature of food means a functional food should be or should

look like a traditional food. More than half the definitions (15/26) mentioned the nature of the

food concept. Third, according to half of the selected definitions (13/26), the concept of function

refers to benefits beyond its basic nutritional functions that make it a functional food. The last

identified concept was regular consumption, which means a functional food must be part of a

normal diet or fit a normal consumption pattern in a specific geographic and/or cultural context,

mentioned in 9 out of 26 of the selected definitions. Therefore, a food which is defined as a

functional food in one country may not necessarily be also considered as a functional food in

another country. With an extensive literature review and elicitation of experts‘ opinions, Doyon

20

and Labrecque (2008) generated a working definition: ―a functional food is, or appears similar

to, a conventional food. It is part of a standard diet and is consumed on a regular basis, in

normal quantities. It has proven health benefits that reduce the risk of specific chronic diseases

or beneficially affect target functions beyond its basic nutritional functions‖ (p1144).

Several of the 26 different definitions summarized by Doyon and Labrecque (2008) refer

to the disease prevention or treatment functions of functional foods. For example, DeFelice

(2007) defined functional foods as any substance that is a food or part of a food that provides

medical and/or health benefits, including the prevention and treatment of disease. An article

from CSIRO Human Nutrition (2004) defined functional foods as: ―Foods that may be eaten

regularly as part of a normal diet, that have been designed specifically to provide a

physiological or medical benefit by regulating body functions to protect against or retard the

progression of diseases such as coronary heart disease, cancer, hypertension, diabetes and

osteoporosis‖. These definitions are closely related to the disease prevention claims examined in

this study, although as has been noted, there is supportive but not conclusive scientific evidence

for the role of functional foods in the prevention or treatment of chronic diseases. In the current

North American functional foods market, only function claims and disease risk reduction claims

are permissible. Disease prevention claims are regulated as drug claims and not allowed on food

products in either Canada or the United States.

2.2 Developments in Functional Food Markets

Functional food is reportedly one of the fastest-growing sectors in the food industry.

Fitzpatrick (2003) estimated that the Canadian functional food market was expanding by 7% to

10% each year. Estimates of the size of the global functional foods market vary widely, from

21

US$11 billion to $155 billion annually, depending on the source of the information and on the

definitions of a functional food (Weststrate, Poppel and Verschuren, 2002). The global market

for functional foods was estimated to be worth about US$33 billion in 2000 (Hilliam, 2000), and

had grown to an estimated of US$85 billion by 2006 (Nutrition Business Journal, 2007).

The Canadian retail market for the functional food and nutraceutical industry was estimated

to be about US$2.5 billion in 2005 (AAFC News Release, 2007) 3

. Currently, the largest and

most rapidly expanding functional food and nutraceutical market in the world is in the United

States (World Nutraceuticals, 2006). In 2003, Canadian international trade in functional food and

nutraceuticals represented an estimated 3% of the global market compared to the United States

(35%) and EU (32%) (D‘Innocenzo, 2006). The United States, Japan and the European Union

are the three largest importers of Canadian functional food and nutraceutical products (Statistics

Canada, 2003).

Functional dairy products, such as Omega-3 milk, probiotic and prebiotic4 yogurts, are

one of the earliest and most widely adopted forms of functional foods. The European market for

functional food is dominated by digestive health products, with dairy products originally being

the key sales category at an estimated US$1.35 billion as early as 1999 (Hilliam, 2000). Some

observers have noted a 30-50% price premium in high volume functional food segments, such

3 Nutraceutical, a product isolated or purified from food that is generally sold in medicinal forms not usually associated with

food, and has been demonstrated to have a physiological benefit or provide protection against chronic disease (Health Canada).

http://www.hc-sc.gc.ca/sr-sr/biotech/about-apropos/gloss-eng.php#n (Accessed Dec.1, 2010) In the Canadian regulatory context,

nutraceuticual is classed as a ―Natural Health Product‖. Note that nutraceuticals and functional foods are substitute products

from a health benefits point of view, and they are often mentioned simultaneously in the existing literature.

4 Probiotics are defined as ―live microorganisms which, when administered in adequate amounts confer a health benefit on the

host.‖(FAO/WHO, 2001) Prebiotics are ―non-digestible substances that provide a beneficial physiological effect on the host by

selectively stimulating the favorable growth or activity of a limited number of indigenous bacteria.‖ (Reid et al., 2003)

22

as functional dairy products in Europe (while raising questions about the sustainability of those

premia in the long-run without proven efficacy and credible health claims) (Menrad, 2003).

Canadian functional foods companies produce a wide range of functional food products.

The highest profile products are from agricultural and fisheries companies which produce high

quality fatty acids and oil products that contain Omega-3-fatty acids (Basu, Thomas and

Acharya, 2007). It has been observed that Canada has the potential to expand its production of

functional foods:

―the country boasts more than 300 companies - from small start-ups to multinational

enterprises - many that are internationally recognized for their bioactive ingredients, such as

soluble fibre from oats, barley and pulses; omega-3 fatty acids from fish and flax oil;

unsaturated fatty acids from canola oil; plant sterols and stanols from vegetable oils; and protein

from soy. Industry is also interested in incorporating into food products functional ingredients,

such as probiotic bacterial cultures; prebiotics (e.g. fructo-oligosaccharides) from corn;

bioactives concentrated from berries and flax; and novel fibres from pulses‖ (Agriculture and

Agri-Food Canada, 2009).

More than half of the above bioactive ingredients cannot form the basis of a health claim in

food labels or in advertisements under current Canadian regulations. It seems that Canadian

functional food companies might have incentives to lobby the government to approve more

health claims for functional foods in the Canadian food market. In a survey, Statistics Canada

(2007) asked Canadian functional food and nutraceutical companies about the Canadian

regulatory environment for health claims: 60% of the respondents believed that government

verification of health claims would lead to higher sales within the industry (Tebbens, 2005).

Thus, the regulatory agency is likely to face ongoing pressure to expand allowable health claims

in the Canadian market, which needs to be balanced with the imperative of consumer protection.

Figure 2.1 is derived from a survey of food companies by Statistics Canada in 2007, it

shows the number of Canadian companies with functional and health products in the market by

23

targeted disease and/or condition. It suggests that the Canadian functional food industry is an

emerging industry and focuses on products targeting cardiovascular disease, energy, immune

system, gut health, mental ability, cancer and diabetes. The disease-related distribution is

relatively narrow focusing on a few major health concerns.

Figure 2.1 Number of Canadian Companies with Functional

and Health Products in the Market

Sources: Functional Foods and Nutraceuticals Survey, Statistics Canada (2007).

In the next section, the notion of information asymmetry in the context of functional foods

is explored. Information asymmetry underpins consumer research on functional food choices

given the credence nature of the health benefits.

2.3 Functional Foods and Information Asymmetry

According to information theory, in general goods are categorized as having search,

experience and credence quality attributes (Nelson, 1970 and Darby and Karni, 1973). The

quality of a search attribute is determined before purchase (e.g. the appearance of the food); an

24

experience attribute is one whose quality is determined after purchase (e.g. taste); while the

quality of a credence attribute cannot be determined either before or after purchase (e.g. the

nutritional component of the food). Functional foods are usually considered to be foods with

credence attributes, since (in the absence of labelling) the functional properties cannot be

inferred by the consumer before or even after purchase (Zou and Hobbs, 2006).

The market for functional foods is characterized by information asymmetry for consumers

with respect to the presence, level and efficacy of the functional attributes. This point has been

recognised in the literature: Menrad et al. (2000) argue that information and communication

activities are needed to deal with the problem that consumers have limited information and

knowledge about the health effect of some functional ingredients; Malla, Hobbs and Perger

(2005) discussed labelling as a solution to information asymmetry regarding health attributes in

food products.

Akerlof‘s (1970) well-known paper on the market for lemons underpins most subsequent

work on information asymmetry as a cause of market failure. Briefly, he argued that good

quality products could be eventually driven out of the market by poor quality products because

of information asymmetry between buyers and sellers. Sellers have more knowledge about

product quality; if the quality cannot be credibly signalled, goods of both qualities will sell at

the same price. The absence of a price premium for higher quality results in only low quality

products being offered for sale.

Two information problems are apparent in the functional foods market. One is whether the

functional component is present, such as the presence of elevated levels of Omega-3 fatty acids.

Of course, this is related to different quality levels of functional foods: higher quality (functional

25

foods) and lower quality (no specific health benefit: ‗normal‘ foods). The other information

problem is whether the functional health claim is credible, which depends on whether

consumers believe that consumption of functional foods will lead to specific health outcomes.

If consumers cannot identify functional foods, they will not be willing to pay a higher

price for foods with those qualities, and the economic incentive to produce a functional food is

not present. Eventually there would be no functional food in the market. Clearly, producers of

functional food have a strong incentive to label their products. Labelling can be viewed as a

means of providing information for consumers to make choices, and there is considerable debate

about the merits of different labelling policies. Unfettered, unregulated labelling can lead to

misleading health claims, confusion and distrust among consumers, while the absence of

labelling information or very restrictive, limited labelling information does little to alleviate the

uncertainty and information asymmetry facing consumers. The two types of information

asymmetry problem in the functional foods market are related to the credibility of labelling.

Food labelling is generally classified as mandatory or voluntary labelling. Mandatory

labelling requires that products with certain kinds of ingredients be labelled clearly (CFIA,

2001); for example, food containing allergens in Canada, or food containing genetically

modified (GM) ingredients in the European Union. Mandatory labelling is usually imposed

where it is perceived that the private market would fail to provide sufficient information to

consumers. Voluntary labelling allows firms the choice of labelling an attribute, but if the

products are labelled, the information must be true and credible (Caswell, 2000); this approach

is used for voluntary labelling of GM-free foods in Canada and the US. For functional foods

with a ‗health benefit‘ attribute, voluntary labelling is appropriate as firms have an incentive to

26

label the presence of a positive attribute, albeit labelling claims may be regulated to ensure

credibility and prevent misleading health claims.

Relative to the labelling of functional foods, the labelling of GM foods has been studied

much more widely. The GM food labelling literature provides some useful insights for potential

research directions with respect to functional foods. Hu, Veeman and Adamowicz (2005)

suggest that there are two research areas in terms of the impact of (GM) labelling. One area is to

study existing (GM) food labelling and how this labelling affects consumer behaviour; the other

area is related to the labelling context and how different types of labelling information affect

consumers‘ choices. A considerable literature exists on the effect of food labelling information

on consumer choices; researchers have used stated choice methods (e.g. Blend and Ravenswaay,

1999), contingent valuation (e.g. Roosen, Lusk and Fox, 2003), and experimental auctions (e.g.

Rousu et al., 2004). Since functional food is a relatively new food category in Canada, and the

regulatory environment for functional food labelling claims remains opaque, the second area of

the labelling context is an appropriate focus for current research, particularly with respect to

health claim attributes and understanding how consumers make choices in an uncertain

information environment.

Definitions of functional foods vary, as do estimates of the size of the market for functional

foods. Nevertheless, the inherent information asymmetry given the credence nature of functional

food is indisputable. It is also not difficult to conclude that labelling (in whatever form) is a

central means of providing consumers with information about functional foods. The next section

reviews previous studies on Canadian consumer attitudes towards functional foods.

27

2.4 Canadian Consumer Attitudes towards Functional Foods: Previous Studies

A growing literature has examined consumers‘ attitudes towards functional foods

consumption in recent years (for example, Hailu et al., 2009; Lusk and Parker, 2009; Verbeke,

2005; Sorenson and Bogue, 2005; Cox, Koster and Russell, 2004 and DeJong et al., 2003).

Consumers‘ attitudes and beliefs about the relationship between food and health play an

important role in their acceptance of functional food products (Gilbert, 2000; West et al., 2002

and Labreque et al., 2006). Health and nutrition information could change consumers‘ attitudes

towards functional foods, specifically, reading health claims on food labels can enhance

consumer perceptions of functional foods (Bech-Larsen and Grunert, 2003).

Chase et al. (2007) examine Canadian consumer purchasing behaviour and attitudes

towards Omega-3 products using ACNielsen HomescanTM

data combined with ACNielsen

Panel TrackTM

survey data from 2003 to 2005 with a sample of 9,825 respondents in Canada.

The study examined consumers‘ purchases of four types of Omega-3 products: eggs, yogurt,

milk and margarine. This study is interesting because it uses data on actual purchases of Omega-

3 products. Using a Probit model, the author found that the most frequent purchaser of Omega-3

products in Canada is baby boomers (born 1946-1955); while, the presence of children in the

household could increase the purchasing frequency of Omega-3 yogurt and Omega-3 margarine;

and the important factors that could affect the purchase of Omega-3 products were reading the

Nutrition Facts panel and health benefits. This study also found that the growth in sales of

Omega-3 products (eggs, milk, yogurt and margarine) exceeded the growth in sales of

conventional food products during the study period from 2003 to 2005 in the Canadian food

market. Interestingly, the researchers suggested that the approved health claims for certain

28

specific Omega-3 fatty acids by the FDA in the United States was a step in the right direction,

while such a claim has not yet been approved in Canada.

Peng, West and Wang (2006) conducted research on consumer attitudes towards and

acceptance of Conjugated Linoleic Acid (CLA) enriched dairy products in Western Canada.

Unlike the above analysis of Omega-3 products, consumers‘ demand for CLA-enriched

products is not observable since they are not yet available in the market. A telephone survey was

conducted in 2004 with a sample of 803 respondents (401 in Alberta and 402 in British

Columbia). Seven CLA-enriched dairy products were examined in this study, which were 1%,

2%, whole and flavoured CLA milk, CLA yogurt, CLA butter and CLA cheese. Using a

contingent valuation method, consumers were asked to provide ordinal ranking preferences for

those seven products. A Probit model was applied for the final data analysis, with the

calculation of marginal effects. The results showed that consumers believed that food choices

played a role in preventing chronic diseases and this perception had an impact on their CLA

yogurt and CLA cheese choices. The study also asked consumers to rate the extent to which

they believed the five out of seven types of risk reduction claims permitted by Health Canada as

described in Chapter 1. The results indicated that consumers who were more likely to believe

health claims of food labels and the potential health benefits of CLA were more interested in

purchasing CLA products. The authors also concluded that baby boomers should be targeted as

CLA dairy product consumers, while the presence of children also had strong impacts on

consumers‘ purchase intentions for some CLA dairy products. However, previous purchase

experience of Omega-3 products and soy products was not related to interest in CLA milk

products. The authors indicated that perhaps there were few ―spill-over‖ effects between

29

different categories of functional food products and suggested that future research should

address consumer willingness to pay premiums for CLA dairy products.

Hailu et al. (2009) used conjoint analysis to examine consumer evaluation of functional

foods and nutraceuticals in Canada. The paper applied a choice experiment to examine three

products enriched with probiotics: yogurt, ice cream (functional food) and a pill (nutraceutical).

Four attributes were selected and tested in this study. They were mode of delivery (three

examined products), health claims (general well-being claim and risk reduction claim for cancer),

source of claim (government, manufacturer and other organization) and price. Data was collected

through a mall intercept survey with a sample of 322 respondents in 2005 in Guelph, Ontario. A

Multinomial Logit model and cluster analysis was applied to the data. The results showed that

consumers place a high value on claims verified by government, but little value on ‗non-verified‘

claims made by product manufacturers. Health claims were the key to an effective marketing

strategy for functional foods and nutraceuticals. The authors also found that there were clear sub-

groups of consumers, some of which prefer delivery of probiotics as a nutraceutical and others

who prefer to consume it in the form of a food.

To further understand what types of research methodologies are suitable for the study

consumers‘ functional food choices, the following section provides an overview of consumer

research methodologies.

2.5 Review of Consumer Research Methodologies

2.5.1 Stated Preference Methods versus Revealed Preference Methods

The research methodology adopted in this study is one of the Stated Preference Methods: a

choice experiment. Broadly speaking, there are two basic economic valuation methods to

30

examine individual‘s preferences for market and non-market goods: Revealed and Stated

Preference Methods. Revealed Preference Methods, one of the earliest consumer demand

theories, pioneered by Paul Samuelson (1938), states that consumers‘ preferences can be

revealed by their purchasing habits. This method is usually applied to the demand for existing

products where data is available on consumers‘ actual choice behaviour in the real market.

Stated Preference Methods are usefully empirical research techniques to understand and

predict the decision maker‘s choice behaviour among discrete products, and were originally

developed in marketing research in the 1970s. The methods focus on assessing consumer

responses to potential product characteristics (Quagrainie, Unterschultz and Veeman, 1998).

Kroes and Sheldon (1988) state that the term ‗stated preference methods‘ refers to a family of

techniques which use individual respondent‘s statements about their preferences in a set of

options to estimate a utility function.

An extensive literature discusses and compares Stated and Revealed Preference Methods

(see for example, Kroes and Sheldon, 1988; Adamowicz, Louviere and Williams, 1994 and

Morikawa, Ben-Akiva and McFadden, 2002). There are two major advantages of Stated

Preference Methods over Revealed Preference Methods. First, Stated Preference Methods allow

researchers to estimate and predict the demand for new products with non-existing attributes in

a situation where the revealed preference data is not yet available. Second, the data collected by

Revealed Preference Methods, the attributes and attribute levels of non-market goods (for

example a lake for recreational fishing) usually do not have variability over time in a single

cross-section, and it is difficult to estimate the value changes provided by the non-market good

without panel data (Louviere, Hensher and Swait, 2000). For example, the observation of

choosing a location for lake fishing is fixed in a single time period, while the change effects of

31

respondents‘ choices can be observed over multiple periods of time. It is also common to have

collinearity among multiple attributes (e.g. the relationship of sales and price) in revealed

preference data, which makes it difficult to isolate the effect of one variable from another. A

well designed Stated Preference study should avoid these problems.

The challenge of the Stated Preference Method is its hypothetical nature. Consumers may

provide unrealistic statements if there is no cost to over (or under-) stating their willingness to

pay (WTP). Also, if consumers are unfamiliar with the product (which could be the case for a

new functional food attribute), their stated WTP may be inaccurate, because this method asks

respondents to state their WTP values but does not record an actual choice action as is the case

with revealed preference studies. The fact that Stated Preference Methods are based on what

people say rather than what people do, is the source of its greatest strengths and its greatest

weaknesses compared with Revealed Preference.

2.5.2 Contingent Valuation and Conjoint Analysis

Generally speaking, Contingent Valuation and Conjoint Analysis are two major Stated

Preference Methods. The Contingent Valuation Method (CVM) is a survey based economic

technique and has often been applied to non-market goods, such as environmental resources and

wildlife (Hanemann, 1994). CVM uses a survey instrument to ask respondents questions

directly about their WTP or willingness-to-accept (WTA) as a compensation for the non-market

goods in a hypothetical scenario. Conjoint Analysis (CA) estimates the structure of a

consumer‘s preference given his/her overall evaluation of a set of alternatives that are pre-

specified in terms of levels of different attributes (Green and Srinivasan, 1978). Both CVM and

CA are survey-based stated preference approaches for data collection. The major difference

32

between CVM and CA is that CVM focuses on a single scenario to collect the precise

information from each respondent‘s choice, while CA tends to examine a respondent‘s

preference by providing a richer description of the attributes trade-offs in the overall scenarios

and results in a smaller variance for welfare values than CVM (Adamowicz et al., 1998). The

data collection approaches that can be used to conduct conjoint analysis include judgment data

(e.g. rating and ranking conjoint) and choice data (e.g. choice experiment) (Louviere, 1988).

A choice experiment is a type of choice based conjoint analysis, a sub-family of conjoint

analysis techniques, also called a Discrete Choice Conjoint Experiment (Louviere, 1988). It is

consistent with Random Utility Theory and Lancaster‘s (1966) Consumer Demand Theory. The

choice experiment was developed from Conjoint Analysis and Discrete Choice Theory by

Louviere and Woodworth in 1983, allowing the researcher to study the choice process and

attribute trade-offs process simultaneously. The choice experiment method evaluates the values

of attributes of a product by asking people to choose the most preferred product out of a few

available products. In each scenario of a stated choice experiment, the alternative options are

described as combinations of different levels of the attributes, and the descriptions of the

alternatives vary among scenarios. The respondents make their choices in a series of

hypothetical choice sets. The preference effects of attributes can be derived by observing the

changes of respondents‘ choices due to the variation in the choice alternatives.

2.5.3 Application of Choice Experiments in the Food Sector

A number of previous studies have used choice experiments to examine consumers‘

attitudes towards functional foods or foods with credence attributes (West et al., 2002, Cranfield,

Deaton and Shellikeri, 2009; Steiner, Gao and Unterschultz, 2010; Quagrainie, Unterschultz and

33

Veeman, 1998; Larue et al., 2004; Lusk and Parker 2009; Hu, Woods and Bastin, 2009; Hovde

et al., 2007).

West et al. (2002) used a choice experiment to analyze consumers‘ responses to different

types of functional foods, produced by conventional, organic, and GM technology. This is one

of the earliest studies on the functional food sector. A representative sample of 1,008 Canadian

household food shoppers responded to the stated preference experiment administered through a

telephone survey in 2001. Each choice set in the questionnaire asked consumers to choose

between the same foods produced by three different food production processes. As it was a

phone survey, the number of other characteristics describing the foods had to be small, and the

three alternatives differed in terms of price and the presence or absence of a functional health

property. A Mixed Logit model was used to analyse the responses. The authors found that the

majority of respondents were willing to pay a price premium for functional foods, particularly if

the functional ingredients were derived from plants. Consumers were less receptive to a

functional property if the functional food was a meat product. The results also indicated that

many Canadian consumers would avoid GM foods regardless of the presence of functional

health properties, and they are likely to accept conventional and organic functional foods if the

prices are reasonable.

Another example of the application of this method is the paper by Quagrainie, Unterschultz

and Veeman (1998). A stated preference experiment was administered in major cities in western

Canada in 1996 via a mail survey; there were 530 respondents. The research question dealt with

how product origin, packaging, and selected demographics affect consumers‘ choice of red

meats. Several attributes were selected for each different fresh meat product, including price,

product origin and packaging. A Nested Logit model was used to analyse the Stated Preference

34

data. The estimation results indicated that the consumers generally preferred Alberta fresh beef

rather than a more general Canadian origin, but the consumers were indifferent between fresh

pork from Alberta and elsewhere in Canada. Consumers‘ age, household income and family size

were found to have an effect on meat choice.

A recent article by Lusk and Parker (2009) applied a choice-based conjoint experiment to

examine consumer preferences for the amount and type of fat in ground beef. This paper linked

consumers‘ beef choices to their health concerns and fat content. Several attributes and levels

were selected, including the levels of Omega-3 fatty acid, conjugated linoleic acid (CLA) and

saturated fat. The examined product, ground beef enriched with Omega-3 and lower fat content,

is also a type of functional food. The goal was to examine preferences for a ‗heart healthy‘ beef

product. A survey was mailed to 2,000 randomly selected households throughout the United

States in April 2007 with 241 surveys returned, a 12.7% response rate. This paper aimed to

determine consumers‘ willingness to pay for beef with a healthier fat content, and to determine

the importance of fat content in beef relative to other beef attributes. WTP estimates showed

that consumers placed significant value on beef enhanced with Omega-3 fatty acids, ranging

from $1.30 to $2.21 per-pound of ground beef depending on total fat content. The authors

suggested that it might be profitable for the beef industry to market and sell products that are

healthier for the consumer (e.g. heart healthy beef). This paper confirmed the importance of

Omega-3 functional products. The method used to calculate the total WTP is also helpful to

develop the WTP estimates in this study.

Hu, Woods and Bastin (2009) used the choice experiment method to study consumers‘

acceptance and willingness to pay for blueberry products with nonconventional attributes:

organic, Kentucky-grown and sugar-free. An in-store intercept survey was conducted in

35

Kentucky with a sample of 557 respondents in 2007. The results found strong evidence that

demographic variables had a significant impact on consumers‘ preferences. For example,

consumers with different age, household income and years of education have different

preferences. Depending on their personal characteristics, consumers‘ preferences and

willingness to pay differ for various attributes. For example, younger and mid-aged consumers

with low to moderate income valued the attribute Kentucky-grown much higher than the organic

feature for a pure blueberry jam product. The analysis in this study used the WTP method to

obtain the marginal values when the selected attributes also contained demographic interaction

variables in a Mixed Logit model. The calculation of marginal WTP when there is significant

demographic interaction effects involved in the estimation model provides insights that are

valuable for the estimation methods used in this study.

Hovde et al. (2007) use a choice experiment to identify market preferences for high

selenium beef in the United States. The survey design included three attributes: price premium,

health claims and origin. Health claims levels included the FDA level A and FDA level C

claims. Data was obtained from a nationwide internet survey with a sample of 1,304

respondents in 2006. A Multinominal Logit Model was estimated. Unexpected results showed

that respondents did not prefer the high-selenium beef products with the FDA level A and C

health claims. The authors explained that because the words ―cancer‖ and ―selenium‖ were

included in the claims; both words might have elicited negative perceptions about the product.

Also, consumers were unfamiliar with the function of the new functional ingredient, selenium,

which might reduce the risk of certain cancers. One interesting finding was that those with less

health-oriented lifestyles, including those who did not exercise and who use tobacco, preferred

36

high-selenium beef with health claims. Those findings provide some insights for this study, such

as the importance of careful wording in the examined health claims.

The choice experiment studies summarized here represent only a fraction of the literature

in the area of consumers‘ attitudes towards and beliefs about functional foods or healthy food

choices. The next chapter (Chapter 3) focuses on how a choice experiment was applied in this

study to examine the effects of labelling on Canadian consumers‘ choices of Omega-3 milk

products. Chapter 3 discusses the selection of product attributes and levels, the choice

experiment design and the survey and data collection.

37

Chapter 3: Research Methodology: the Choice Experiment

The choice experiment approach is a widely adopted empirical methodology in the

economics literature (Louviere, Hensher and Swait, 2000; Hensher, Rose and Green, 2005;

Adamowicz et al., 1998 and Hu, Veeman and Adamowicz, 2005). For an individual consumer‘s

choice problem, the classic random utility approach of consumer theory is appropriate (Manski,

1977). The choice experiment method is consistent with Random Utility Theory. It is a data

generation approach which depends on the design of choice tasks to reveal factors influencing

choices and to understand how respondents make choice decisions (Louviere, Hensher and

Swait, 2000). A choice experiment is used to observe the effects upon one variable, a response

variable, given the manipulation of the levels of one or more other variables in the choice sets

(Hensher, Rose and Green, 2005). The choice set is a subset of all alternatives in a universal set

that are available at the time of the choice and have a non-zero probability of being chosen

(Adamowicz et al., 1998).

The first step in conducting a choice experiment is to refine the understanding of the

problem being studied and ask the question: ―what does the research hope to achieve?‖ After

understanding the problem, the researchers identify a list of alternatives, attributes and attribute

levels which are appropriate for the choice experiment. This step is called stimulus refinement,

which means brainstorming and then narrowing the range of alternatives to consider in the

experiment (Hensher, Rose and Green, 2005). The key issues in designing a choice experiment

method include selecting the attributes and levels, the experimental design and the treatment of

the no-choice option. These aspects of choice experiment design are discussed in sections 3.1,

3.2, and 3.3, respectively. The chapter concludes with a description of the survey design and

data collection.

38

3.1 The Selection of Attributes and Levels in the Choice Experiment

The choice experiment explores how consumers value and make trade-offs among the

selected attributes. The selected attributes need to properly reflect the competitive environment

of the available alternatives and/or be closely relevant to consumers‘ decision making (Blamey,

Louviere, and Bennett, 2001). For example, as mentioned in Chapter 1, the research questions of

this study focus on examining consumers‘ response to full labelling, partial labelling and the

verification of health claims, when making functional food choices. Therefore, full labelling

(function claims, risk reduction claims and disease prevention claims), partial labelling (symbol)

and verification organizations (government and third party verification) were selected as the

main attributes for inclusion in the choice experiment.

When selecting attributes, the following aspects need to be considered by analysts. First,

although many products might have the potential to become examined product(s), only the most

representative product or service should be selected. The selected product or service will be

used as the alternatives in the choice sets. For example, this study focuses on the effects of

labelling on consumers‘ functional food choices, so the examined product should be a

representative functional food which can credibly carry the labelling attributes. Second, not all

attributes and levels that are relevant to consumers‘ choice decisions for the examined product

are included in the choice experiment design. For example, taste might be a significant attribute

in consumers‘ food choices. However, if taste itself is not relevant to the research question it

should not be included as an attribute in the experiment.

In some studies, the utility brought by the levels of a single attribute is referred to as

part-worth utility or marginal utility. The more levels are brought in by analysts, the more

complexity of choice tasks or tradeoffs that are added. It is important to thoroughly understand

39

the research topic and simplify the selected attributes and levels. The information on the

relevant attributes and levels could be obtained through the use of focus groups (Lindlof and

Taylor, 2002), discussion with relevant industry stakeholders, and review of relevant literature.

For this study, the process of selecting attributes included interviewing individual consumers

and scientific experts, informal market research on dairy products and a review of relevant

literature.

Omega-3 milk has been selected as the product examined in this study. Why is Omega-3

milk chosen? First, the dairy sector is the earliest and a dominant functional food sector, and

milk is widely consumed among Canadians. Second, Omega-3 fatty acid is one of the most

popular functional ingredients as it is found in several functional food products. According to the

World Health Report of the World Health Organization (2002), the potential health benefits of

Omega-3 include cardiovascular disease prevention and control, cancer prevention, improved

immune function and brain health. Cancer and cardiovascular/heart disease are also two out of

the four leading causes of death in Canada. Third, health claims are not currently permitted on

food products enriched with Omega-3 in Canada. Fourth, as Health Canada continues to review

health claims in recent years, there may be changes in the near future regarding the regulation of

health claims. For example, explicit health claims for products enhanced with Omega-3

ingredients may be allowed in the future.

Table 3.1 summarizes the selected attributes and levels in this choice experiment. Health

claims, verification organization, heart symbol, Omega-3 functional ingredient and price are the

selected attributes. The labelling effects include partial labelling and full labelling, which are

included as two separate attributes in this design. Three types of full labelling health claims are

defined as function claims, risk reduction claims and disease prevention claims. Function claims

40

and risk reduction claims have already been applied on functional foods in the U.S. and

Canadian markets. Function claims, such as ―calcium could build strong bones‖, usually do not

need to be pre-approved in the United States, but certain conditions need to be met to be applied

on food labels in Canada. Risk reduction claims (which link a nutrient to a particular disease)

need to be supported by scientific evidence and need to be pre-approved in both Canada and the

United States. Disease prevention claims are regulated as drug claims and are not yet permitted

on any food products in either Canada or the United States.

Table 3.1: Attributes and Levels in Choice Experiment

ATTRIBUTES LEVEL 1 LEVEL 2 LEVEL 3 LEVEL 4

1. Omega-3

Contains

Omega-3

(included as a

alternative

specific

constant)

None

- -

2. Health

Claims

(full

labelling)

Function

Claim:

―Good for your

heart.‖

Risk Reduction Claim:

―Reduces the risks of

heart disease and cancer‖

Disease Prevention

Claim:

―Helps to prevent

Coronary Heart

Disease and

Cancer―

None

3. Symbol

(partial

labelling)

Heart Symbol

None - -

4. Verification

Organization Government:

Health Canada

Third Party:

Heart and Stroke

Foundation of Canada

None

-

5. Price

(per two-litre)

$1.99

(conventional)

$2.69

(conventional or

Omega-3)

$3.59

(conventional or

Omega-3)

$4.49

(Omega-3)

In this study, the full labelling health claims attribute has four levels: Function Claim

(FC), Risk Reduction Claim (RRC), Disease Prevention Claim (DPC) and no claim. The partial

41

labelling attribute is represented by the ‗heart symbol‘ defined as two levels: with a heart

symbol and without. The research design also explored the role of verification organizations in

lending credibility to a health claim, with ‗Verification Organization‘ also included as an

attribute. Three levels are specified for the verification organization attribute: Government

(Health Canada), Thirty Party (Heart and Stroke Foundation) and none.

The Omega-3 attribute enters the choice experiment as an alternative specific constant

variable, which has two levels: containing Omega-3 and none. In other words, each choice set

contains two Omega-3 milk options as the first two alternatives and one conventional milk

option as the third alternative. Although Omega-3 could have been used as a separate attribute in

the experiment, initial tests of experimental designs showed that its inclusion required too many

restrictions with respect to the relationship with other attributes and significantly reduced the

efficiency of the design. For example, in the case of the absence of any Omega-3 in the choice

set, all the other attributes (except price) would have to be restricted to the ―none‖ levels (e.g. no

health claim, no symbol and no verification).

The price attribute is directly related to the measurement of consumers‘ WTP for other

attributes in the experiment. The selected price levels should reflect but not be limited to the

retail price in the market, with the price range covering the likely lowest price and the highest

price for milk products as much as possible. According to observations for milk prices from food

markets in Saskatoon, Edmonton, Calgary, Toronto, Vancouver and from information gathered

from the internet, the four selected levels of the price attribute are: $1.99, $2.69, $3.59 and $4.49

for a two-litre carton of milk. Note that the highest price ($4.49) is applied only to Omega-3 milk

and the lowest price ($1.99) is only applied to conventional milk to maintain the realism of the

choice task.

42

There are four alternatives in each choice set: the first two are Omega-3 milk, the third is

conventional milk and the fourth is the no-purchase alternative (discussed in section 3.3). There

are some restrictions for the attribute combination appearing in choice alternatives. Health claims

and the heart symbol can only appear in the functional milk alternatives since conventional milk

cannot contain a health claim or heart symbol. Verification organizations can only appear with

health claims, otherwise there is nothing to be verified by those organizations. The Omega-3

attribute cannot appear in the conventional milk and the no-purchase alternatives, but it must

appear in the first two alternatives in each choice set to indicate that they are the functional food

alternatives.

Although Omega-3 milk products already exist in the Canadian functional food market,

health claims for Omega-3 are not currently permitted, therefore a choice experiment allowed the

potential health claims, and verification of those claims by different organizations, to be

examined. The design of the choice experiment is discussed in the next section.

3.2 Choice Experiment Design

An efficient experimental design is used to maximize the information collected from the

stated preference choice experiment. The objective is to create efficient choice sets, including

how to combine attribute levels into product profiles and how to put profiles into choice sets

(Batsell and Louviere, 1991).

Two more concepts need to be clarified before outlining the formal design used in this study.

One issue concerns main effects and interaction effects, and the other deals with labelled

experiments and unlabelled experiments. The main effect is the independent effect of a particular

treatment upon the dependent variable, the choice. The measurement of the main effect is by the

43

estimated parameter of that treatment variable. An interaction effect is the combined effect of

two or more treatments upon the dependent variable. The interaction effect could be measured by

the estimated parameter of the combined variables. The main effects and interaction effects

determine the degrees of freedom of the experiment, which is directly related to the design of the

minimum number of profiles. The number of the profiles needs to be sufficient to estimate both

the main effects and interaction effects.

A labelled experiment refers to an experiment using a label or a brand title for each

alternative of the choice task. For example, brand names for milk products, such as Dairyland or

Beatrice could be used as a label for an alternative. An unlabelled experiment refers to an

experiment using a generic title for each alternative and the generic title does not convey any

information to the decision maker. A label or a brand title in a labelled experiment has a brand

effect in the design. The design processes are different for labelled experiments and unlabelled

experiments. Generally speaking, to make an efficient design, the unlabelled experiment requires

a smaller number of profiles than the labelled experiment. The design of this study uses an

unlabelled experiment since a brand effect is not directly relevant to the research questions being

examined.

A full factorial design is a design in which each level of each attribute is combined with

every level of all other attributes (Louviere, Hensher and Swait, 2000). It allows all possible

combinations of the attribute levels and alternatives to be used, and allows all the main effects

and interaction effects to be estimated. The number of possible profiles in a full factorial design

is equal to LA for an unlabelled experiment and L

MA for a labelled experiment, where L is the

number of levels, A is the number of attributes and M is the number of alternatives (Hensher,

Rose and Greene, 2005). Usually it is too costly and tedious to expose respondents to all possible

44

choice sets. There are some strategies available to reduce the number of choice sets given to

respondents, including reducing the number of levels, using a fractional factorial design,

blocking the design or using a fractional factorial design combined with a blocking strategy

(Hensher, Rose and Greene, 2005).

A fractional factorial design is a design in which only a fraction of all treatment

combinations is used in the choice sets. How are the combinations chosen? Two principles

should be considered which are orthogonality and balance (Hensher, Rose and Greene, 2005).

An orthogonal design requires all attributes to be statistically independent of each other, which

means zero correlations between attributes. A balanced design requires the probability of each

attribute level occurring equally often for each attribute of each alternative in all choice sets. One

way to express a fractional factorial design could be LA-P

, where L and A have the same meaning

as above, and P represents the size of the reduced fraction of the full factorial design determined

by researchers (Holmes and Adamowicz, 2003). There is a minimum design requirement that the

number of the profiles in a fractional factorial design must satisfy the estimation of the main

effects and the two-way interaction effects. The limitation of the fractional factorial design is that

it cannot guarantee to impart the maximum amount of information about the parameters of the

attributes relevant to each specific choice task (Hensher, Rose and Greene, 2005).

Related to the orthogonal fractional factorial design, the optimal or statistically efficient

design is another design method in the choice experiment literature and was the method adopted

in this study. The development or application of the optimal efficient design method is discussed

in Kuhfeld, Tobias and Garratt (1994); Huber and Zwerina (1996); and Kanninen (2002). The

optimal efficient design needs to have balance, orthogonality and minimal level overlap (Huber

and Zwerina, 1996). The goal of the optimal efficient design is to minimize the variance and

45

covariance of parameter estimates and maintain those optimal design properties as much as

possible (Kuhfeld, Tobias and Garratt, 1994). The criterion of the optimal efficient design is to

have the design be as efficient as possible, which means to search for a minimum-variance

design. Such a design is known as a D-optimal design, which aims to maximize the determinant

of the variance-covariance matrix of the model to be estimated. There are some differences

between the orthogonal fractional factorial design and the optimal efficient design. The

orthogonal fractional factorial design aims to make the attributes of the design statistically

independent and uncorrelated. In contrast, the optimal efficient design optimizes the amount of

information in the design and aims to be statistically efficient but is likely to have correlation

among attributes (Hensher, Rose and Greene, 2005).

The design of the choice experiment in this study is based on Warren F. Kuhfeld‘s (2005)

Marketing Research Methods in SAS and Johnson et al. (2007). This is an unlabelled generic

design because there are no ‗brands‘ in the alternatives for the examined products. There are four

alternatives in each choice set. The first two alternatives are Omega-3 functional milk; the third

alternative is conventional milk; and the fourth alternative is the ‗no purchase‘ option.

In this study, the choice set design is somewhat different from other choice experimental

studies. In most studies, only one representative product needs to be selected, and the difference

among alternatives in each choice set is the changes of the treatment combinations, which is a

relatively straightforward choice design. For this partial constant design, two products are

included in each choice set to mirror reality: conventional milk taken as the base/reference

product, and Omega-3 milk examining additional functional attributes. In the real milk market,

conventional milk still dominates the market and functional milk, such as Omega-3 milk, only

takes a relatively small market share. In this experimental design, it is therefore reasonable to

46

have conventional milk as the base/reference product, so respondents could make trade-offs not

only between different treatment combinations, but also between conventional milk and

functional milk.

Technically, there are two methods whereby conventional milk can be incorporated into the

experimental design. One straightforward method is to take conventional milk as a possible

treatment combination/profile, and in this type of design conventional milk could occur in any

alternative in each choice set. This study applies the other method, by taking conventional milk

as a fixed alternative in a partial-constant design. The design treats conventional milk as a partial

constant alternative with only the price level changing. Thus, all other attributes remain constant

and stay at the ‗none‘ level (i.e. just conventional milk without health claims or verification).

This is very similar to the creation of the ‗no-purchase‘ option. The ‗no-purchase‘ option is

explained in detail in the next section. A pilot study was used to investigate both design

alternatives. Feedback from respondents and the experiment results favoured the second method

which treats conventional milk as a partial constant alternative. This structure provides

respondents with a more realistic choice task, making comparisons among profiles and products

easier. The remainder of this section explains the creation of this design.

Applying the SAS Macro program (Kuhfeld, 2005), the %Mktruns and %Mktex macros

were used to create initial candidate choice sets, and then the %choiceff macro was used to

select the optimal design with the highest D-efficiency score5 out of 100 designs. This study

contains four attributes (with 4/3/2/4 levels), and there would be 96 choice sets for a full factorial

design. Therefore, the optimal design needs to include an appropriate amount of choice tasks to

satisfy a complex model which can handle the possible main effects and interaction effects, the

5 D-efficiency score is defined as: 100*(1/N|X’X

-1|

1/K) (Kuhfeld, 1997).

47

possible alternative specific effects and demographic effects. The design also needs to be

practical since it will be tedious and might be hard for respondents to efficiently complete too

many choice tasks within one survey.

The number of minimum profiles for this study can be calculated by the number of main

effects and two-way interaction effects. Evaluating the number of degrees of freedom is

important to estimate the entire set of main effects and interaction effects (Louviere, Hensher,

and Swait, 2000). This study includes four attributes with 4/3/2/4 levels. Each main effect is

associated with L-1 degrees of freedom (L is the number of levels for each attribute). Therefore,

there are 9 degrees of freedom for the main effects. This design also considers partial interaction

effects between partial labelling and full labelling, and between full labelling and verification

organizations. For interaction effects, the degrees of freedom are calculated by (L1-1)*(L2-1),

where L1 and L2 are the levels of the two-way interacted attributes. For this study, there are 9

degrees of freedom for the interaction effects. Therefore, there are 18 degrees of freedom for

both main effects and interaction effects for one alternative in the choice sets. Since there are two

alternatives in each choice set that involves both main effects and interaction effects (Omega-3

milk options), in total there are 36 degrees of freedom in this study. The number of profiles in the

choice design must exceed the number of degrees of freedom. Thus, the minimum number of

profiles for this study is 36. This design includes 48 profiles which exceeds this requirement.

Finally, it was necessary to impose some restrictions when generating the profiles. Based

on retail prices for existing Omega-3 milk products, prices in the first two Omega-3 alternatives

cannot realistically be the lowest price $1.99, and conventional milk in the third alternative

should not realistically be at the highest price $4.49. If there is no health claim in a product, there

should not be a verification organization in that product. It is impossible to expect each

48

respondent to complete the full 48 choice sets used in this design. Accordingly, the %Mktblock

macro in SAS was used to block the design into six blocks, so that each respondent faced eight

choice sets. Eight was considered a reasonable choice set number for each respondent, given that

there were other questions that needed to be answered in the survey. An example of a choice set

is shown in Table 3.2.

49

Table 3.2 An Example of a Choice Set from the Survey

Attributes Option A Option B Option C Option D

Symbol on

Package

I would not

purchase

any of these

milk

products.

Additional

Ingredient

Health

Claims

Verifying

Organization

of Health

Claims

Price $3.59 $4.49 $2.69

Choices □

□ □

Option A is a 2-litre carton milk enriched with Omega-3. It has a health claim “Reduces the Risks of

Heart Disease and Cancer”, verified by the Heart & Stroke Foundation. This carton of milk costs

$3.59.

Option B is a 2-litre carton milk enriched with Omega-3. It has a health claim “Helps to Prevent

Coronary Heart Disease and Cancer”, verified by Health Canada. This carton of milk costs $4.49.

Option C is a carton of 2-litre conventional milk and costs $2.69.

Option D can be selected if you would not buy options A, B, or C.

In this example, option B was chosen.

50

3.3. The treatment of the „No-purchase‟ Option

The ‗No-purchase‘ option is a constant alternative in each choice set and does not vary

among choice sets. The purpose of the no-purchase option is to simulate the real purchase

environment where consumers always have the option to choose not to buy a product. The

inclusion of the no-purchase option in the choice set avoids forced choices. If the no-purchase

option is not provided in the choice experiment consumers are forced to make choices among the

hypothetical alternatives which might change the values of attributes relative to a real market

situation. Forced choices could also bias the product demand estimation and WTP results

(Carson et al., 1994).

There are several formats through which to present a no-choice option in the choice

experiment: ‗no-purchase‘ and ‗the current purchased product‘, etc. A no-purchase option is

often applied when the research objective is to simulate realism or measure market demand.

Another alternative is to allow consumers to choose their current purchased product. This

enables a researcher to examine which attribute or level of the new product is attractive to induce

consumers to switch from their current purchased product (Batsell and Louviere, 1991). There is

no clear guideline on which format should be applied, and is primarily a matter of choice for the

researcher. For this study, the no-purchase option has been applied as a constant alternative in

each choice set to simulate the real shopping environment.

One popular approach in the literature to deal with the inclusion of the no-purchase option in

the choice set is by the measurement of an alternative specific constant (ASC) variable to

normalize the alternative utilities (Train, 2003). The ASC approach is used in this study. This

method treats the no-purchase option similarly to other alternatives and assumes the marginal

utilities of the attributes in the no-purchase option to be zero since there are no attributes

51

associated with this option. The ASC approach measures the relative value of the utility

associated with one alternative compared with another alternative. The difference in the utility

associated with the no-purchase option and other purchase options is captured by the ASC

variable. The coefficient of the ASC variable for the no-purchase option is a relative utility

associated with that option relative to the other purchase options.

This section has explained some key steps in the design of the choice experiment,

including the selection of attributes and levels, the experimental design and the treatment of the

no-purchase option. The following section provides details on the remainder of the survey and

the data collection process used in this study.

3.4 Survey and Data Collection

The choice experiment was administrated through a consumer survey6

. The survey

contained six sections. The first section asked about respondents‘ general milk consumption

habits and previous experience consuming conventional milk and Omega-3 milk. This section

was used to collect data on reference points for different attributes, which is discussed in Chapter

6. The second section asked respondents to complete a series of choice tasks, which was the

primary source of choice data for the estimation models in this thesis. The third section gathered

data on consumers‘ trust in food labels, symbols and health claims, and the source of health

information for food products. The fourth section asked about consumers‘ attitudes, beliefs and

consumption of functional foods and regular foods. The fifth section gathered information on

consumers‘ health problems and histories, their health status and exercise habits. The final

6 This survey was approved by the Behavioural Research Ethics Board at the University of Saskatchewan in January

2009.

52

section of the survey contained some socio-demographic information. A copy of the entire

survey is provided in Appendix 4.

Before conducting the formal survey a pilot study was completed in March 2009 in

Saskatoon. This pilot survey was carried out in different locations in Saskatoon - at the Midtown

Plaza Mall, the Lawson Heights Mall, the student centre at the University of Saskatchewan and

various Tim Hortons Café locations. Forty two respondents completed a pilot survey. This pilot

study provided very useful feedback in finalizing the survey instrument. For example, in the pilot

survey, fat percentage was initially used as a milk product attribute in the choice experiment.

However, it was found that fat percentage is a dominant factor in determining milk choices for

many respondents but the fat percentage has little relationship with the research questions and it

clouded the analysis and created bias. After careful consideration this attribute was removed

from the choice experiment design and taken instead as a constant attribute (i.e. in introducing

each choice set, respondents were told to assume that all fat percentages of milk products are

available). The pilot study also assisted in refining the wording of questions and removing

unnecessary questions.

The formal survey was conducted in July 2009 as an online survey in English with

respondents recruited from across Canada, with the exception of Quebec due to resource

limitations. A professional marketing research company administered the online survey and

recruited participants using survey sample recruitment firms. As an incentive to participate in the

survey, participants had a chance to win a monthly sweepstakes which could be redeemed for

cash or they could choose to donate any winnings to assigned charities. The survey was kept

open until the target sample size had been reached. A total of 924 completed surveys were

53

received, with 740 of these responses usable. The procedures used to identify qualified responses

are explained below.

Several quality control procedures were implemented to control for the quality of survey

data. For example, first, two screener questions were used to screen out unqualified participants,

which were ‗do you consume cow‘s milk at least once per month?‘ and ‗are you one of the

primary food grocery shopper in your household?‘. Participants answering ‗No‘ to either

question did not proceed with the survey. Second, the data from respondents who completed the

survey in 10 minutes or less were not used in the analysis. From the pilot study it was clear that it

should usually take 15-20 minutes to complete the survey and individuals who finished it within

ten minutes might not take the survey seriously (See appendix 3 for details). There were 47 (or

5.1%) respondents who took less than 10 minutes to complete the survey in the initial 924

respondents.

Furthermore, for the choice experiment section initially 8 choice sets had been randomly

assigned to each respondent. To ensure the quality of the collected data, two more choice sets

were included. These were ‗trap‘ choice sets, designed to weed out respondents who were

answering the choice sets inconsistently. ―Trap question‖ was originally developed from

Sociology literature (e.g. Barry and Bateman, 1996). In this study, one trap choice set was to

include a strictly dominated alternative (e.g. one alternative included superior attributes and the

lowest price, so this option was a logical answer if a respondent was answering rationally). This

was positioned to be the ninth choice set for each respondent for simplicity. Only data from those

respondents answering this choice set logically were used in the analysis. An additional trap

choice set (the tenth choice set) was an identical repeat of an earlier choice set to check whether

a respondent‘s answers are consistent. If a respondent‘s answer was different in the repeated

54

choice set from his/her earlier choice, his/her choice data were considered invalid and were

removed from the data used for analysis. For ease of programming and data analysis, the

repeated choice set was the same within each block (e.g. the second choice task was always

repeated as the tenth choice task for each respondent). These two additional ‗trap‘ choice sets

were used as flag questions to enhance the quality of the final choice data. In total, there were 95

(or 10.3%) respondents out of the initial 924 respondents who failed to pass the two flag

questions. Finally, to avoid an ordering effect in the choice tasks, the order of the choice sets was

randomized for each respondent in this survey with the exception of the additional ninth and

tenth trap choice sets.

As mentioned above, some unqualified data were not used in the formal analysis in this

study. Those unqualified data include those respondents who completed the survey too quickly

and who did not pass the ―trap‖ questions. In Appendix 3, some post hoc estimation are

examined by using the unqualified data. The results in the appendix show that the

unused/unqualified data set is problematic and may provide inconsistent and not robust

estimation results. Therefore, it is reasonable for this study to remove those ―noise‖ data from the

total data set and use the qualified data sample in the formal analysis. To the researcher‘s

knowledge, this type of work has been rarely done in the literature. The findings in Appendix 3

should have some useful implications to improve data quality and produce valid measurements

in terms of web survey methodology (see Appendix 3 for details).

This chapter has described the design of the choice experiment and administration of the

survey in this study. The next three chapters present the analyses of the survey data. Chapter 4

presents the econometric models and the descriptive analysis of the survey data. Chapter 5

presents estimation results for the effects of labelling using seven different discrete choice

55

models. Chapter 6 examines the existence of reference-dependent effects in Canadian

consumers‘ functional food choices.

56

Chapter 4: Econometric Models and Descriptive Analysis

This chapter focuses on developing the econometric models and estimation methods to

examine the effects of labelling on Canadian consumers‘ functional food choices. As described

in Chapter 1, this study examines Canadian consumers‘ responses to different types of health

claims, symbols, credibility of verification organizations, and the effects of other attitudinal and

behavioural factors when making functional food choices. Furthermore, the analysis examines

how different types of labelling information affect consumers‘ preferences for functional foods.

This chapter also presents the descriptive analysis from the online consumer survey which

includes a sample of 740 respondents (see Chapter 3 for details). This chapter contains two sub-

sections: a discussion of the econometric models and estimation methods, and a descriptive data

analysis and details of the factor analysis used to examine the behavioural and attitudinal

variables.

4.1. Econometric Models and Estimation Methods

According to McFadden (1974) and Train (2003), an individual i receives utility U when

choosing an alternative j with a group of attributes Xij from a choice set. The utility has been

modelled with two components: an observed deterministic component Vij and an unobserved

stochastic component εij of the utility function. The utility received from alternative j is

represented by:

Uij = Vij + εij (4.1)

Where Vij = f(Xij), the deterministic component, is a function of the attributes of the alternatives.

In the choice model, individual i faces a choice of one alternative from a finite choice set C. The

niP

57

probability Pij that alternative j will be chosen equals the probability that the utility gained from

this choice is no less than the utility of choosing another alternative in the finite choice set. The

probability of individual i choosing alternative j is expressed as:

(4.2)

In this study, consumer i faces the choice of one alternative among Omega-3 milk,

conventional milk and the no-purchase option, given various attribute level combinations in each

choice set. The probability of consumer i choosing alternative j equals the probability that the

utility received from alternative j is greater or equal to the utility when choosing any other milk

alternative or the no-purchase option.

McFadden (1974) developed the Conditional Logit model to estimate these probabilities

assuming the stochastic error term is independent and follows a Type-I extreme value

distribution. Assume the observed deterministic component Vij is a linear function of perceived

product attributes Xj, so Vij = β’Xj. The choice probability of consumer i choosing alternative j in

the Conditional Logit Model is formed as:

(4.3)

where µ is a scale parameter which is usually assumed to be 1, β is a vector of parameters, and k

is an index representing the chosen product by consumers from the choice set (k =1,…,K, where

K = 4 in this study). Parameters in this model can be estimated by the maximum likelihood

estimation method.

K

k

k

j

ij

X

XP

1

)'exp(

)'exp(

},&;VV{Pr ikij CkkforjobP ikijij

58

This study uses Random Utility Theory as the underlying theoretical framework. The

indirect utility of consumer i choosing alternative j can be expressed as the following function:

Uij = Xijß + ej (4.4)

where ß is a vector of estimated parameters and Xij represents a vector of the selected

attribute levels in the choice set, and ej is the error term associated with the utility brought by

alternative j, which cannot captured by the attributes. Given the specified attributes and levels of

milk products in this study, a linear additive indirect utility function of consumer i choosing

alternative j in one choice set is specified as:

Uij = ß1NoPurchase +ej (j = no purchase) (4.5)

Uij = (1-NoPurchase)*(ß2FunctionClaimj + ß3RiskReductionClaimj +

ß4DiseasePreventionClaimj + ß5HeartSymbolj + ß6Govermentj + ß7ThirdPartyj

+ß8Omega-3j + ß9Pricej) + ej (j ≠ no purchase) (4.6)

This is a base model which specifies the utility function with the main effects variables in

the choice experiment. As specified in Table 3.1 in Chapter 3, four attributes are included in this

choice experiment, which are health claims, verification organizations, heart symbol and price.

Those attributes include different levels which are all dummy-coded, and they become the main

variables to test the effects of labelling in the random utility function. The health claim attribute

(full labelling) is separated as four dummy variables: Function Claim (FC), Risk Reduction

Claim (RRC), Disease Prevention Claim (DPC) and No Health Claim (the base level).

‗HeartSymbolj‘ is a dummy variable representing partial labelling equal to 1 if the milk product

has a red heart symbol, otherwise equal to 0. The verification organization attribute is

represented by three dummy variables: Government Verification (Government), Third Party

59

Verification (Heart and Stroke Fundation) and No Verification (the base level). ‗Omega-3j‘ is an

alternative specific constant attribute equal to 1 if the milk product includes Omega-3 ingredient

and equal to 0, otherwise. ‗NoPurchase‘ is a dummy variable equal to 1 if an alternative

represents the option of not purchasing any of the milk products, and equal to 0 otherwise.

‗Pricej‘ is the retail price of the milk product in alternative j. Note that for each selected attribute,

the base level has been removed from the utility function (4.6) to avoid the singularity problem

with dummy variables.

When designing the choice experiment as presented in Chapter 3, the possible interaction

effects between the main variables was also considered. The indirect utility function of consumer

i choosing alternative j with both the main effects and the interaction effects could be expressed

as follows:

Uij = (1-NoPurchase)*(ß2FCj + ß3RRCj + ß4DPCj + ß5HeartSymbolj + ß6GOVj+ ß7TPj +

ß8Omega-3j + ß9Pricej + ß10FCj*Heartj + ß11RRCj*Heartj + ß12DPCj*Heartj

+ß13RRCj*GOVj + ß14RRCj*TPj + ß15DPCj*GOVj + ß16DPCj*TPj) + ej (4.7)

Where FCj*Heartj, RRCj*Heartj, and DPCj*Heartj are the interaction variables between the

variable Heart Symbol (partial labelling) and the variables Function Claim (FC), Risk Reduction

Claim (RRC) and Disease Prevention Claim (DPC) (full labelling). RRCj*GOVj is the

interaction effect between the variables Risk Reduction Claim and Government Verification

(GOV). RRCj*TPj is the interaction effect between the variables Risk Reduction Claim and

Third Party Verification (TP). Similarly, DPCj*GOVj and DPCj*TPj are the interaction effects

between the variables Disease Prevention Claim and Government Verification and between

Disease Prevention Claim and Third Party Verification.

60

Equation (4.7) includes both the main effects and the interaction effects. A main effect

represents the effect of one level of the selected attribute on the dependent variable measured

independently from other variables. A two-way interaction effect is the first order interaction

between two main variables. An interaction effect between two variables exists if consumers‘

preferences for the levels of one attribute depend on the levels of the other (Louviere, Hensher

and Swait, 2000). In this study, the potentially relevant interaction effects are between partial

labelling and full labelling, and also between health claims and verification organizations.

Consumers‘ preferences for three types of health claims (full labelling) might depend on whether

a heart symbol (partial labelling) is present on food labels. For example, consumers might be less

sensitive to the presence of a heart symbol when the risk reduction claim is also present on food

labels, compared with the presence of a heart symbol combined with the function claim. The risk

reduction claim itself might convey enough health information to consumers no matter whether

or not a heart symbol is present, while this might not be the case for the function claim. Similarly,

consumers‘ preferences for each type of health claim might depend on which organization

verifies the health claim. For example, consumers might prefer the government to verify a risk

reduction claim relative to a third party verified risk reduction claim.

In addition to the choice experiment section, the other parts of the survey also collected

extensive information on consumers‘ attitudes towards and beliefs about functional food

consumption, as well as socio-demographic information. One method to incorporate the

attitudinal information into the utility model is through the interaction between the main

variables and exogenous covariate variables. The indirect utility function in equation (4.8)

incorporates the socio-demographic and attitudinal information into the random utility model:

61

Uij = (1-NoPurchase)*(ß2FCj + ß3RRCj + ß4DPCj + ß5HeartSymbolj + ß6GOVj+ ß7TPj

+ß8Omega-3j + ß9Pricej +γnZn*Xj) +ej (4.8)

Where Zn represents the selected exogenous covariate variables (e.g. income, education,

gender and heart disease); Xj represents the product attributes that can be interacted with Zn; γn

represents the coefficients of those interaction variables. For example, Zn*Xj can be specified as

the interaction variables HeartDiseasei*RRCj, Incomei*Omega-3j, Edui*RRCj or

Genderi*Omega-3j, etc. HeartDiseasei*RRCj is an interaction effect between the variables Risk

Reduction Claim and Heart Disease, and it captures the possibility that consumers who have

heart disease might be more sensitive to the presence of a risk reduction claim on food labels

than those who do not have heart disease. A similar interpretation pertains to the variable

HeartDiseasei*DPCj. Incomei*Omega-3j is an interaction effect between the variables Omega-3

and Income, capturing the effect that the presence of Omega-3 on food labels might receive a

different response from higher income consumers relative to lower income consumers.

Edui*RRCj is an interaction effect between the variables Risk Reduction Claim and Education,

which allows a test of whether consumers‘ responses to the presence of a risk reduction claim are

different between better educated consumers and less educated consumers. Genderi*Omega-3j

captures the interaction between Omega-3 and Gender, to estimate female consumers responses‘

to the presence of Omega-3 on food labels relative to male consumers. Interaction effects

between the main attributes and other exogenous covariates are also possible.

As mentioned previously, the survey collected a great deal of information about

consumers‘ trust in food labels, purchase habits, healthy lifestyle habits, health status, etc. This

information could help to explain consumers‘ responses to choice tasks. However, it is

impossible to directly incorporate all the individual attitudinal variables into the estimation

62

model. Statistically speaking, because some questions in the survey address related issues taking

those questions directly as the covariate variables and interacting them with the main variables

might create serious colinearity problems among these covariate variables and cause endogeneity

between error terms and the explainable variables, which might become a source of bias in the

estimation results. Besides, a large number of covariate variables are not good for model

estimation. Factor Analysis is a popular method for data reduction. The purpose of Factor

Analysis is to examine relationships between variables and find a way to condense or summarize

the information contained in an original set of variables into a smaller set of factors for

subsequent regression, correlation or discriminant data analysis (Hair et al., 1992). Thus,

interactions between key factors and the main variables can be introduced into the utility model.

The key factors derived from Factor Analysis are orthogonal to each other, so there is no

correlation among factors. Factor Analysis was used to capture additional information about

individual preferences and attitudes, and is discussed in more detail below. The indirect utility

function of consumer i choosing alternative j that incorporates three key factors is expressed as:

Uij = (1-NoPurchase)*(ß2FCj + ß3RRCj + ß4DPCj + ß5HeartSymbolj +

ß6GOVj+ ß7TPj +ß8Omega-3j + ß9Pricej + γnZn*Xj +

ß10RRCj*Factor1i + ß11RRCj*Factor2i + ß12RRCj*Factor3i +…) +ej (4.9)

Where RRCj*Factor1i is an interaction effect between Risk Reduction Claim and Factor 1 (the

definition of the factors is discussed below); RRCj*Factor2i captures the interaction between

Risk Reduction Claim and Factor 2; and RRCj*Factor3i is an interaction effect between Risk

Reduction Claim and Factor 3, and so on. Each factor could be viewed as a function of a group

63

of original attitudinal variables, and the meaning of each factor is determined by the variables in

that group. A detailed discussion of the Factor Analysis is provided in section 4.2.2.

A number of discrete choice models are available that differ in the assumptions made about

the distribution of the error term (Train, 2003). For example, the Conditional Logit model‘s error

term is assumed to have a type-I extreme value distribution. The Probit Logit model‘s error term

is assumed to have a normal distribution. The discrete choice models typically used to estimate

consumers‘ choice behaviours, are the Conditional Logit, the Mixed/Random Parameter Logit

model, the Latent Class model, the Nested Logit, and the Ordered Probit model. This study

focuses on estimation results from the Conditional Logit model, the Random Parameter Logit

model and the Latent Class Model.

The Conditional Logit (CL) model is a standard and fundamental starting point from

which to derive other advanced models in the family of discrete choice models. However, the

limitations of the CL model are obvious as well. The major limitation is that the estimated

coefficients of the attributes are fixed to be the mean values of all respondents‘ responses. It

ignores the variation of the estimated coefficients and cannot handle preference heterogeneity

among consumers. Consumer heterogeneity is an important issue in food markets, especially

when dealing with differentiated products, such as functional food products, where target

consumer preferences might be quite different from other consumers. The second major

limitation of the CL model is its well-known IIA assumption or property (independence of

irrelevant alternatives). The IIA property assumes that the ratio of the probability for any two

alternatives is completely independent of the existence and attributes of any other alternatives

(see Ben-Akiva and Lerman, 1985). It assumes that the errors are independently distributed

across alternatives even for repeated choices, which is unrealistic. The CL model cannot avoid

64

the restrictive substitution pattern of the IIA property (Louviere, Hensher and Swait, 2000). It is

often necessary to relax the IIA assumption in practice. However, it is common in the literature

to estimate the CL model and the estimated result serves as a benchmark for other discrete choice

models.

Several different approaches have been developed to address the limitations of the CL

model in the economics literature, such as the Mixed/Random Parameter Logit model, the Latent

Class Model (LCM), Probit model and the Nested Logit model. The Mixed Logit (ML) model is

a popular method to explore the unobserved heterogeneity in choices, as is the LCM. In this

study, significant improvements in model fits were found from the estimation results of both the

ML model and the LCM, compared with the estimation results of the CL model (see Chapter 5

for details).

The Mixed Logit model is very flexible and can approximate any random utility model

(McFadden and Train, 2000). The Mixed Logit model was developed by Boyd and Mellman

(1980), Jain, Vilcassim and Chintagunta (1994), Bhat (1998) and Train (1998), to identify a

broad range of consumers‘ preference heterogeneity. According to Train (2003), the Mixed Logit

probabilities are the integrals of standard logit probabilities over a density of parameters. The

ML model assumes that rather than being fixed, the parameters of attributes follow certain

specific distributions across the respondents in the sample. Specifically, the choice probability of

the Mixed Logit model of individual i choosing alternative j can be expressed as:

(4.10)

Where θ represents the distribution parameters of coefficient β (such as the mean and

covariance of β); Pij is the standard logit probability function given in equation (4.3); coefficients

dfPP ijij )|(

65

β are distributed with the probability density function f (β | θ) defined over a set of parameters θ.

There is usually no closed form for the integration expression for ijP in equation (4.10).

Therefore, its likelihood function cannot be efficiently estimated with Maximum Likelihood

estimation (Hu, Veeman and Adamowicz, 2005). However, the probability function ijP in

equation (4.10) can be estimated by a simulation method over the density function f (β | θ).

According to Train (2003), the procedure for the simulation method includes three steps.

The probabilities for any given value of θ: (1) draw a value of β from the density function f (β |

θ), and name it βr with the superscript r = 1 to represent the first draw; (2) calculate the )( r

ijP

with the logit formula for the first draw; (3) repeat steps 1 and 2 many times (usually more than

100 times), and average the results. The average simulated probability can be expressed as:

(4.11)

Where R is the number of draws; ijP

is an unbiased estimator of ijP , and its variance

decreases as R increases. The summation of ijP

is equal to 1 over alternatives, which is a useful

property for forecasting. The simulated log likelihood function is given by inserting the

simulated probabilities into the log likelihood function as the following question:

(4.12)

)(1

1

rR

R

ijij PR

P

I

i

J

j

ijij PdSLL1 1

ln

66

Where dij is an indicator that dij =1 if individual i chose alternative j and zero otherwise. The

maximum simulated likelihood estimation (MSLE) is derived by maximizing SLL over the value

of θ (Train, 2003).

The Mixed Logit model is a weighted average of the logit formula over different values of

parameters β, and the weights are given by the density function f (β | θ). Actually, the standard

Conditional Logit model is a special case of the Mixed Logit model where the mixing

distribution f (β) is degenerate at fixed parameters b, f(β) =1 for β = b and 0 for β ≠ b (Train,

2003). The mixing distribution f (β) can also be discrete when β takes a finite segment of distinct

values. Suppose β takes S possible segment values labelled as b1,…,bS, with probability Hs when

β = bs. In this case, the Mixed Logit model becomes the Latent Class Model (Train, 2003).

Although the Mixed Logit model explicitly accounts for preference heterogeneity by

allowing estimated parameters to vary randomly over individuals, it might be sufficient and more

interesting to use the LCM to identify consumer heterogeneity by separating individuals into

several classes. Consumers within each class have similar preferences. Following McFadden

(1986), Boxall and Adamowicz (2002) and Shen (2009), the LCM assumes that a discrete

number of classes or segments of respondents are sufficient to account for preference

heterogeneity among classes. The LCM allows the choice attribute data and individual

consumer‘s personal characteristics to simultaneously explain choice behaviour (Boxall and

Adamowicz, 2002). This study also presents LCM estimations results in Chapter 5.

The latent classes capture the unobserved heterogeneity in the population, and each class is

estimated with a different parameter vector. Note that in the Conditional Logit model, the vector

β is not specific to an individual or segment. If we assume the existence of S segments in the

67

population, the choice probability of individual i choosing alternative j in class s of the Latent

Class Model is expressed as:

(4.13)

Equation 4.13 is a standard probability specification of the Conditional Logit model for

class s. The LCM simultaneously estimates the above probability equation and predicts the latent

class probability Hij of individual i being in class s. So the unconditional probability equation of

the LCM is expressed as (Boxall and Adamowicz, 2002):

(4.14)

Another issue in the LCM concerns how to choose the number of classes, S. The Akaike

Information Criterion (AIC) and Bayesian Information Criterion (BIC) are used to decide the

number of classes, S (Louviere, Hensher and Swait, 2000, and Boxall and Adamowicz, 2002).

The criterion indicates that the preferred number of S classes is found where the values of AIC

and BIC are minimized (Louviere, Hensher and Swait, 2000).

The parameters estimated by the Discrete Choice models should not be interpreted by

their absolute values which are not comparable, but must be jointly explained with other linked

estimated parameters (Hensher, Rose and Greene, 2005). Willingness-To-Pay (WTP) is often

adopted by researchers to jointly interpret the estimated parameters and identify the money

values associated with changes in each attribute. The marginal WTP indicates the maximum

K

k

kss

jss

sij

X

XP

1

|

)exp(

)exp(

ij

S

s

sijij HPP |

68

amount that the respondent would be willing to pay in order to receive/avoid a particular

attribute of the product (Burton et al., 2001). The marginal WTP can be derived as follows:

(4.15)

Where βk is the estimated parameter of attribute xk and βprice is the parameter for price,

kxWTP represents the money value that respondents are willing to pay for the attribute of xk of the

product characteristics.

When interaction effects are included in the estimation function, Hu, Woods and Bastin

(2009) suggest calculating marginal values as WTP estimates for the total effect of an attribute.

For example, when considering demographic differences, sometimes interactions between

demographic information and product attributes need to be included in an estimation model. A

general formula to calculate marginal values for WTP estimates is:

(4.16)

where βattribute and βprice are the coefficients of an attribute and the price variable,

respectively; D is a vector of the demographic variables being interacted with product attributes;

βD is the vector of corresponding coefficients associated with interaction terms; and βD* D

appears when the attribute or price are used as interaction variables (Hu, Woods and Bastin,

2009). Notice that equation 4.15 actually is a special case of equation 4.16 when term βD* D is

equal to zero.

price

kxk

WTP

price

Dattribute DWTP

*

69

This section has specified the econometric models and estimation methods used in this

study. Sections 4.2 and 4.3, which follow, focus on the descriptive analysis of the survey data

and present the Factor Analysis.

4.2 Descriptive Analysis

This section contains the descriptive analysis of sample characteristics and attitudinal

characteristics for the survey data.

4.2.1 Sample Characteristics

The online survey was administered to respondents recruited from across Canada (with the

exception of Quebec) in July 2009. As discussed in Chapter 3, 740 usable responses were

received. Table 4.1 compares the sample in this study with the Canadian Census data on the basis

of demographics. The sample was reasonably representative with respect to age, size of

household, and number of children in the household. Females were over represented in the

sample at 67.8% compared with 51.1% in the Canadian population. Recall that a screener

question ensured that only those who were the primary household food shoppers were included

in the sample, which likely explains the high proportion of female respondents. The mean values

for household income and education were slightly higher than the Canadian population, which is

to be expected with an internet-based survey.

70

Table 4.1 Socio-Demographic Characteristics

of the Survey Participants and the Canadian Population

Demographic Characteristics Sample Canadian Population

Age 49.3 45.7a

Gender (Female) 67.8% 51.1%a

Size of Household 2.49 2.49

Number of Children under 18 in

household

0.64 0.55

Income (mean) b $58,953 $41,401

Education

High school or less 35.3% 56.7%

College certificate or trade diploma 40.7% 36% c

University Bachelors degree 19.1%

University Masters degree or higher 5% 7.1%

Source: Statistics Canada 2008 (Census 2006). a. Canadian population between 19-80 the same as the

age range represented in the sample. b. average mean income values of each category, see figure 4.1 (b)

for an income comparison between the sample and the Canadian population by categories. c. combination

of university bachelors degree and college certificate.

Figure 4.1 compares the sample and the Canadian population on the basis of a number of

socio-demographic variables: age, household income, and location. As shown in figure 4.1 (b)

for the specific category comparisons, on average, the demographic distributions show that the

sample data slightly over represent higher-income individuals. The sample in this study contains

a higher percentage of older respondents (more than 45 years old) and a lower percentage of

younger individuals (less than 45 years old). The respondents of this sample were randomly

selected from all Canadian provinces with the exception of Quebec since the survey was only

conducted in English. The sample represents a higher percentage of respondents from western

Canada (British Columbia, Alberta, Saskatchewan and Manitoba), and a lower percentage of

residents from Ontario and the Maritimes. Compared with the national data, with the above

caveats noted, the data collected in this survey appears to be a reasonably representative sample

of English speaking Canadians.

71

Figure 4.1 A Comparison of Socio-Demographic Distributions

between the Sample and the Canadian Population

(a) Age

0%

5%

10%

15%

20%

25%

30%

35%

18-24 yr 25-34 yr 35-44 yr 45-54 yr 55-64 yr 65 yr +

National

Sample

Source: Statistics Canada (2008) and Survey Data

(b) Income

0%

5%

10%

15%

20%

25%

30%

Under $20,000

$20,000 -$39,999

$40,000 -$59,999

$60,000 -$79,999

$80,000+

National

Sample

Source: Statistics Canada (2008) and Survey Data

72

(c) Location

0%5%

10%15%20%25%30%35%40%45%

Sample National Source: Statistics Canada (2008) and Survey Data

4.2.2 Attitudinal Characteristics

The survey collected data on respondents‘ attitudes towards sources of health information

and frequency of consumption of functional foods. This section describes the responses to these

questions. Figure 4.27 represents the health information sources used by Canadian consumers and

their trust in those health information sources. Figure 4.3 ranks the top three health information

sources used by the respondents.

7 Note that figure 4.2 cannot be used to make comparisons between light coloured bars (source of health information)

and black bars (trust in those information sources). Instead comparisons should be made across categories for each

of the two coloured bars.

73

Figure 4.2: Health Information Sources

and Trust in those Information Sources (% of Respondents, n=740)

The question underlying figure 4.2 (dark bars) was: ―Where do you typically get health

information about food products? (Please check all that apply)‖. This is a ―yes‖ or ―no‖ question

for each information source. Figure 4.2 (dark bars) shows the percentage of respondents

answering ―yes‖ to each information source. The results in figure 4.2 (dark bars) shows that

Canadian consumers‘ most frequently used information source is ‗food labels‘, with about 58.9%

of respondents typically getting health information about food products from ‗food labels‘. This

result affirms the importance of the research topic in this study: the effects of labelling on

functional food choices. Canadian consumers‘ second most frequently used source of health

information was the internet (46.9%), while, the third highest was the media (42.4%), such as TV,

74

newspaper, radio, and magazines. Notice that only 7.6% of respondents claim that their health

information source was ‗food companies/industry‘.

The light coloured bars in figure 4.2 summarize responses to the question, ―How much do

you trust the following sources for accurate health information?‖ The original survey used seven-

point Likert scales from ―Don‘t trust at all (1)‖ to ―Completely trust (7)‖. To simplify the results,

respondents‘ answers have been re-coded into a ―0 or 1‖ format, where low levels of trust (1-3)

were coded as ―0‖, and higher levels of trust (4-7) were coded as ―1‖. Figure 4.2 (light coloured

bars) shows the percentage of respondents who identified the information source as highly

trustworthy. The result also indicates that the top three trusted health information sources are: (1)

professionals, such as doctors and nutritionists (94.3%); (2) health organizations, such as the

Heart and Stroke Foundation of Canada (90.1%); and (3) food labels (88%). Note that ―food

companies/industry‖ earned the least amount of Canadian consumers‘ trust as a source of health

information for food products (64%).

Generally speaking, if a source of health information is more commonly used, the trust in

that source is usually relatively high. For example, ―food labels‖ is the source of health

information for most respondents, and respondents also claim a higher level of trust in food

labels (within the top 3). However, a conflicting result is found for ―internet‖ and ―media‖.

Although ―internet‖ and ―media‖ are frequently used health information sources, consumers have

much lower levels of trust in those two sources compared with other information sources. This

suggests that consumers regard the internet and media are less reliable but easily accessible

information sources.

75

As mentioned above, food companies/industry receives the lowest score both for being a

frequently used health information source and for the trust in that information source. However,

notice that ―food labels‖ are also made by food companies, which receives the highest score both

for consumers‘ trust levels and being a frequently used health information source. The reason is

unclear but it might indicate that consumers believe the health information carried by food labels

is subject to regulatory verification by agencies, such as Health Canada. Therefore, the role of

verification organizations in assuring the credibility of the health information on food labels

could be an important factor in consumers‘ food choice decisions. The discrete choice

experiment allows a closer examination of these potential relationships.

Figure 4.3: Top Ranked Health Information Sources (% of respondents, n=740)

The survey included the following question: ―Please rank the top 3 sources you use most

frequently‖. Figure 4.3 presents the top ranked health information sources. The top three health

76

information sources that Canadian consumers indicate they use most frequently are: (1) Food

labels, (2) Professionals, such as doctors and nutritionists, and (3) Media, such as TV, newspaper,

radio and magazines. The dark bars represent the percentage of respondents who rank that source

as the top 1 most frequently used information source. The grey bars represent the percentage of

respondents who rank that source within the top 3 most frequently used information sources. A

few differences are worth noting. For example, about 15% of respondents rank health

professionals as their number one information source, higher than the media (11.5%), while only

26% of respondents rank health professionals within their top 3 most frequently used information

source, whereas 29.5% of respondents ranked media within their top 3 information source.

However, food labels receive the highest response rates for both the number one information

source or within the top three, indicating the importance of food labels to consumers‘ healthy

food choice decisions.

The survey gathered data on the frequency with which respondents consumed different

types of functional foods. Figure 4.4 displays this data. Among functional food products, vitamin

enriched juice is consumed most frequently, followed by mineral enriched cereals and calcium

enriched milk. Probiotic yogurt is also a relatively popular functional food product among survey

respondents. Among the Omega-3 functional food products, Omega-3 eggs are consumed most

frequently, followed by Omega-3 yogurt, Omega-3 milk and Omega-3 cheese.

77

Figure 4.4: The Frequency of Canadian Consumers‟ Functional Food Consumption

(% of respondents, n=740)

0%

10%

20%

30%

40%

50%

60%

70%

Omega-3

Milk

Calcium

Enriched Milk

Omega-3

Eggs

Omega-3

Cheese

Omega-3

Yogurt

Probiotic

Yogurt

Vitamin

Enriched Juice

Mineral

Enriched Cereals

Never Consume Tried Once or Twice Occasionally Consume Regularly Consume

Few respondents appear to consume Omega-3 milk, certainly fewer than calcium enriched

milk, which likely reflects the availability of those two products in the Canadian marketplace,

plus the fact that both function claims and risk reduction claims are allowed on calcium enriched

products, whereas no health claim is yet allowed for Omega-3 functional foods in Canada.

According to the data collected in the survey, on average a typical respondent household

purchases 4.7 litres of milk in one week.

This sub-section has analyzed several characteristics of the sample population. The

following section condenses the attitudinal and behavioural data gathered in the survey into a

smaller set of factors for subsequent regression and data analysis.

78

4.3 Factor Analysis

Factor Analysis is a multivariate statistical method that allows data reduction and

summarization. The purpose of Factor Analysis is to examine the relationships between variables

and to find a way to condense or summarize the information contained in those original variables

into a smaller set of factors for subsequent regression, correlation or discrimnant data analysis

(Hair et al., 1992). Each factor can be treated as a dependent variable for a function of a group of

original variables.

Boxall and Adamowicz (2002) incorporated Factor Analysis into a latent class approach

to examine people‘s selection of wilderness recreation parks. Their paper illustrates a method of

applying Factor Analysis within a latent class model to examine the heterogeneities among

different groups of respondents, which is adopted in chapter 5 of this study (see section 5.7 for

details). In general, Factor Analysis includes several key steps: 1) selecting variables related to

the research problem and deriving the correlation matrix, 2) conducting common factor analysis

and component analysis, 3) the rotation of factors, 4) determining the criteria for the number of

factors to be extracted and criteria for the significance of factor loadings and, 5) naming factors

and how to use factor scores (Hair et al., 1992).

In this study, Factor Analysis was conducted using SPSS. The first step in Factor Analysis

involves choosing the variables related to the research problem. This study is about the effects of

labelling on consumers‘ functional food choices, and the purpose of using Factor Analysis is to

summarize variables and reduce data. In addition to the choice experiment, the survey gathered

data on trust in health claims and labels, consumers‘ attitudes towards functional foods and their

consumption behaviours, health status and demographic information. Although the data collected

in the choice experiment is the primary vehicle for examining labelling effects, Factor Analysis

79

enables the inclusion of summarized attitudinal and behavioural variables to explain the choice

behaviours. In this study, socio-demographic variables were not included in the Factor Analysis

because the objective is to examine whether the attitudinal information could help explain

respondents‘ choice behaviours. Socio-demographic variables are examined separately.

To obtain the factor solutions, a factor matrix needs to be computed based on the

calculation of a correlation matrix and the total variance of the variables. The computation of the

factors determines the best linear combination of variables which account for more variance in

the whole dataset than any other linear combination of variables. The first factor is the single

best linear combination of variables among the dataset, and the second factor is the second best

linear combination of variables subject to the constraint of being orthogonal to the first factor. To

be orthogonal, the second factor must be derived from the remaining variance after the first

factor has been extracted. In other words, the second factor could be viewed as the linear

combination of variables which accounts for the most residual variance after the first factor‘s

effect has been removed from the dataset (Hair et al., 1992). The same method is applied to

derive the subsequent factors until the total variance has been removed. Taking the obtained

factors as the columns and the original variables as the rows, a factor matrix has been

constructed as shown in Table 4.2. The numbers inside the factor matrix are so-called ‗factor

loadings‘ which represent the correlation between the original variables and the factors. Factor

loadings have a range of [-1,+1] and significance levels to measure the relationship between the

original variables and the factors.

One commonly used criterion for factor extraction is the Latent Root (Eigenvalue)

Criterion. Eigenvalue is the sum of squares for a column factor in the factor matrix, which

represents the amount of variance accounted for by a factor. The larger the eigenvalue of a factor

80

the more important is that factor. In Factor Analysis, only the factors having an eigenvalue

greater than 1 are considered significant and all factors having an eigenvalue less than 1 are

considered insignificant and disregarded (Hair et al, 1992). According to the Latent Root

(Eigenvalue) criterion, in the factor matrix in Table 4.2, three factors are considered as the

key/significant factors. The criterion for the significance of factor loadings is simple. A rule of

thumb which has been used frequently as a preliminary examination of the factor matrix is as

follows: factor loadings of +0.5 or more are considered to be very significant (Hair et al., 1992).

In this study, the variables with factor loadings greater than +0.6 are selected as the main

examined variables for each factor.

Factor scores can be computed for each factor if the analyst wishes to create an entirely

new and smaller number of composite variables to replace the original set of variables (Hair et al,

1992). The original raw data and the factor loadings are utilized to compute factor scores for

each individual respondent. The factor score of one factor for one individual is computed by

multiplying the raw data of that individual with the corresponding factor loadings of the

examined variables in the given factor and summing them together. Therefore, an individual who

scores high on several variables that have high loadings for a factor obtains a high factor score

on that factor.

81

Table 4.2: The Key/Component Factors in the Factor Analysis

Factor 1:

Attitudes

towards

Functional Foods

Factor 2:

Trust in Health

Claims and

Nutrition Labels

Factor 3:

Health

Knowledge

Q19B: I trust nutrition labels on food products. .126 .835*** .151

Q19C: The health claims on food products are

accurate.

.268 .836*** .025

Q19D: I trust new food products. .168 .798*** .036

Q24A: Functional foods can maintain overall

wellbeing and improve long-term health.

.898*** .225 .077

Q24B: Functional foods may reduce the risk of

certain chronic diseases.

.918*** .192 .112

Q24C: Functional foods may prevent certain

diseases.

.894*** .183 .093

Q24D: Functional foods are necessary for a healthy

diet and should be consumed regularly.

.874*** .140 .006

Q24F: I am knowledgeable about health and

nutrition.

.099 .131 .902***

Q24G: My friends/relatives ask me for health or

nutrition advice.

.063 .037 .913***

Note: *** represents factor loadings that are +0.6 or more.

Table 4.2 shows that three factors were significant. Factor 1 captures respondents‘

attitudes towards functional foods, which is defined as ―Attitudes towards Functional Foods‖.

The ‗positive‘ attitudes towards functional foods are hereafter captured by the beliefs that

functional foods can maintain wellbeing and can reduce the risk of chronic disease or prevent

certain diseases. The ‗negative‘ attitudes towards functional foods are hereafter captured by the

beliefs that functional foods cannot maintain wellbeing or reduce the risk of chronic disease or

prevent certain diseases. Factor 2 measures consumers‘ trust in health claims, nutrition labels and

new food products. Factor 3 measures health knowledge, capturing respondents who are

82

knowledgeable about health and nutrition (e.g. their friends ask them for health and nutrition

advice). In Factor Analysis, Cronbach‘s Alpha represents a coefficient of reliability and usually

used to measure internal consistency of questionnaire (i.e. how closely related a set of items are

as a group) (Cronbach, 1951). The criterion of using Cronbach‘s Alpha scores to represent

factor‘s reliability is 0.7 (Cronbach, 1951). The alpha reliability scores of this analysis are 0.934,

0.803 and 0.799 for factor 1, 2, and 3, respectively, which captures a high reliability (internal

consistency) by using different original variables to represent each factor. These factors are

included in the econometric analysis discussed in the next chapter.

This chapter explained the econometric models and estimation methods used in this study.

It also included the descriptive analysis of the survey data and presented the Factor Analysis. The

estimation results of the econometric models are presented in Chapter 5 using the various

estimation methods and discrete choice models discussed within this chapter.

83

Chapter 5: Results: the Effects of Labelling on Functional Food Choices

This chapter presents 7 sets of estimation results for the utility models developed in

Chapter 4 to answer the following research questions: (1) what do Canadian consumers prefer:

full labelling or partial labelling? If full labelling, what type of claim do they prefer? (2) how do

they respond to the assurances of different verification organizations: government or third party?

(3) to what extent do consumers value a specific functional ingredient (e.g. Omega-3), and (4)

how do individual characteristics, such as the respondents‘ attitude, trust, knowledge and socio-

demographic factors, influence their functional food choices. The 7 discrete choice models

examine the responses of consumers to different product labelling strategies taking account of

heterogeneity in consumer attitudes. In particular, some of the models recognize that different

consumers make different choices depending on their current health status, the credibility of

health claims, and the extent to which they believe that a change in a functional ingredient will

lead to an improvement in their health status.

The models are: (1) the Conditional Logit model and Willingness-To-Pay (WTP) for the

base model; (2) the Conditional Logit model with interaction effects between main variables; (3)

the Conditional Logit Model with interaction effects of covariate variables and key factors; (4)

the Mixed/Random Parameter Logit model and WTP for the base model; (5) the Latent Class

Model for the base model; (6) the Latent Class Model with interaction effects of covariate

variables and key factors; and (7) the Latent Class Model with class membership indicators.

Table 5.1 describes the variables used in the seven models.

84

Table 5.1 Summary of the Variables for the Estimation Models

Attributes

Abbreviation Description

Price Price The price adopted in the choice experiment for a two-litre

carton of milk, ranging from $1.99 to $4.49.

Function Claim FC = 1 if the milk product has a function claim (‗Good for your

heart‘), otherwise 0.

Risk Reduction Claim RRC =1 if the milk product has a risk reduction claim (‗Reduces the

risks of heart disease and cancer‘), otherwise 0.

Disease Prevention

Claim

DPC =1 if the milk product has a disease prevention claim (‗Helps

to prevent coronary heart disease and cancer‘), otherwise 0.

Heart Symbol Heart = 1 if the milk product has a red heart symbol, otherwise 0.

Government

Verification

GOV = 1 if the health claim on the milk product is verified by

government (Health Canada), otherwise 0.

Third Party Verification TP = 1 if the health claim on the milk product is verified by a

third party (Heart and Stroke Foundation), otherwise 0.

Omega-3 OMG3 = 1 if the milk product contains Omega-3, otherwise 0.

No Purchase NoPurchase =1 if the alternative is not to purchase any milk product in the

choice set, otherwise 0.

Interaction Variables

( between main attributes) FC*Heart - = 1 if the milk product has both a function claim and a red

heart symbol, otherwise 0.

RRC*Heart - = 1 if the milk product has both a risk reduction claim and a

red heart symbol, otherwise 0.

DPC*Heart - = 1 if the milk product has both a disease prevention claim and

a red heart symbol, otherwise 0.

RRC*GOV - = 1 if the milk product has both a risk reduction claim and a

government verification for this claim, otherwise 0.

RRC*TP - = 1 if the milk product has both a risk reduction claim and a

third party verification for this claim, otherwise 0.

DPC*GOV - = 1 if the milk product has both a disease prevention claim and

85

a government verification for this claim, otherwise 0.

DPC*TP - = 1 if the milk product has both a disease prevention claim and

a third party verification for this claim, otherwise 0.

Exogenous Covariate Variables and Key Factors

Heart Disease HeartDisease =1 if a respondent self-reported that he or she has heart

disease, otherwise 0 (6.5% respondents claim they have heart

disease in this sample).

Income Income Mean point of the income in 8 categories (respondents self-

reported annual household income before taxes):

< $25K is coded as $12.5K;

$25K to $49.9K is coded as $37.5K;

$50K to $74.9K is coded as $62.5K;

$75K to $99.9K is coded as $87.5K;

$100K to $124.9K is coded as $112.5K;

$125K to $149.9K is coded as $137.5K;

$150K to $174.5K is coded as $162.5K;

>175K is coded as $187.5K

Note: the ―Income‖ variable is measured by the value of

thousand dollars in all estimation results. Education Edu Respondents self-reported highest education levels (4

categories):

High school or less =0;

College certificate or trade diploma =1;

University Bachelors degree =2;

University Masters degree or higher =3

Gender - =1 if the respondent is female, otherwise 0.

Attitudes towards

functional foods

Attitude Factor 1 of Factor Analysis in Table 4.2

Trust in health claims

and nutrition labels

Trust Factor 2 of Factor Analysis in Table 4.2

Health knowledge Knowledge Factor 3 of Factor Analysis in Table 4.2

Interaction Variables

(between the Main Attributes and the Covariate Variables or the Key Factors)

RRC*HeartDisease - An interaction variable between the variables Reduction Claim

and Heart Disease.

86

DPC*HeartDisease - An interaction variable between the variables Disease

Prevention Claim and Heart Disease.

Omega-3*Income - An interaction variable between the variables Omega-3 and

Income.

RRC*Edu - An interaction variable between the variables Risk Reduction

Claim and Education.

Omega-3*Gender - An interaction variable between the variables Omega-3 and

Gender.

RRC*Attitude - An interaction variable between the variables Risk Reduction

Claim and Attitudes towards Functional Foods (Factor 1).

Omega3*Attitude - An interaction variable between the variables Omega-3 and

Attitudes towards Functional Foods (Factor 1).

RRC*Trust - An interaction variable between the variables Risk Reduction

Claim and Trust in Health Claims and Nutrition Labels

(Factor 2).

GOV*Trust - An interaction variable between the variables Government

Verification and Trust in Health Claims and Nutrition Labels

(Factor 2). Omega3*Knowledge - An interaction variable between the variables Omega-3 and

Health knowledge (Factor 3).

5.1 Base Model: CL Estimates and WTP

In the base model (equation 4.6 in Chapter 4), the effects of the main attributes in the

choice experiment are measured, including full labelling, partial labelling, verification

organization, presence of Omega-3 and Price. Table 5.2 presents the CL and WTP results for the

base model. The value of the Log Likelihood Function is -5275.028 and the Pseudo-R2 is 0.25,

indicating that the goodness of fit of this model is moderately good. All coefficients are

statistically significant at the 1% level and with the expected signs. The WTP values of all

attributes are also statistically significant at 1% and with the expected signs. WTP estimates

87

provide a useful basis on which to compare the relative strength of preferences for the examined

attributes.

Table 5.2: Base Model: CL Estimations and WTP

Variable Coefficient t-ratio WTP

($/2 Litres)

t-ratio

Price -1.461*** -47.618 - -

Function Claim .298*** 2.915 .204*** 2.913

Risk Reduction Claim .679*** 6.652 .465*** 6.694

Disease Prevention Claim .532*** 5.300 .364*** 5.299

Heart Symbol .177*** 3.901 .121*** 3.895

Government Verification .336*** 5.184 .230*** 5.173

Third Party Verification .319*** 4.969 .219*** 4.977

Omega-3 .321*** 3.512 .220*** 3.558

No Purchase -6.12*** -58.395 -4.187*** -66.370

Log Likelihood Function = -5275.028; Pseudo-R2 = 0.25

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

Specifically, the variables capturing full labelling: Function Claim, Risk Reduction Claim

and Disease Prevention Claim, all have positive and significant coefficients which imply that

consumers prefer these three types of health claims to be present on food labels compared with

no health claim (base level). The variable Heart Symbol (partial labelling), has a positive and

significant coefficient, indicating that consumers prefer a heart symbol to be present on Omega-3

milk products, compared with the no symbol (base level). Of interest for this study is which

labelling format consumers are more likely to prefer, partial labelling or full labelling. Since the

88

coefficients of the CL model are marginal utilities which are not comparable, Table 5.2 also

presents the WTP results.

Comparing consumers‘ WTP for full labelling with their WTP for partial labelling in

Table 5.2 shows that, on average, consumers are willing to pay an additional 12 cents for the

heart symbol (partial labelling) for a two-litre carton of milk, compared with no symbol (base

level). However, consumers are willing to pay 20 cents for a function claim relative to no health

claim, which is almost twice the value of their WTP for a heart symbol. Consumers‘ WTP for a

risk reduction claim is 47 cents, almost four times the value of their WTP for a heart symbol (12

cents), while the WTP for a disease prevention claim is 36 cents. The WTP results indicate that

on average, consumers are more likely to prefer full labelling relative to partial labelling, and

within full labelling, consumers are more likely to prefer a risk reduction claim.

The WTP for the variables Government Verification and Third Party Verification are both

positive and significant, which implies that compared to having no verification by any outside

party (the base level), consumers are willing to pay more when the labels include government or

third party verifications of health claims. More specifically, consumers are willing to pay almost

identical amounts: 23 cents extra for the government verification and 22 cents for the third party

verification on a two-litre carton of milk, compared with no verification (base level).

The presence of Omega-3 (an alternative specific constant variable) has a positive and

significant coefficient, indicating that, holding other variables constant, on average consumers

prefer milk products enriched with Omega-3 compared with conventional milk. Consumers are

willing to pay 22 cents more for the presence of Omega-3 per two-litre carton of milk. No

Purchase (an alternative specific constant variable) has a negative and statistically significant

89

coefficient, which implies that not purchasing any milk product in a given choice set, has a

negative impact on consumer‘s utility. The unrealistic negative ‗WTP‘ for No Purchase indicates

how much consumers dislike not purchasing any milk product. Price also has a negative and

significant coefficient which is consistent with economic theory: the higher the product‘s price,

the lower the consumer‘s utility.

Table 5.3 shows the WTP differences between full labelling and partial labelling, and

also between the government and third party verifications. A Wald test is used to test the

significance of WTP differences among attributes. The coefficients of the first three ‗WTP-

difference‘ variables are positive and significant, indicating that consumers prefer a risk

reduction claim relative to a disease prevention claim, they prefer a risk reduction claim relative

to a function claim and they also prefer a disease prevention claim relative to a function claim.

The insignificant coefficients of the last two ‗WTP-difference‘ variables show that there is no

statistical difference in consumers‘ preferences for a function claim and a heart symbol, and

between the two types of verification organizations (Government and Third Party).

Table 5.3: Wald Test for Difference in WTP

Variable Coefficient St. Err. T-ratio Prob.

WTPRRC - WTPDPC 0.101 0.043 2.32 0.020

WTPRRC - WTPFC 0.261 0.044 5.872 0.000

WTPDPC - WTPFC 0.160 0.043 3.704 0.000

WTPFC - WTPHS 0.083 0.074 1.123 0.262

WTPGOV – WTPTP 0.012 0.045 0.257 0.797

Probability from Chi-squared [4] = .000

90

According to the estimation results in Tables 5.2 and 5.3, consumers are more likely to

prefer full labelling relative to partial labelling. Canadian consumers might believe that full

labelling conveys more accurate health information about a product‘s health benefits than partial

labelling. Recall that no statistically significant difference was found between a function claim

and a heart symbol, suggesting that respondents, on average, regard a function claim such as

―Good for your heart‖, as similar to a red heart symbol. Among the three full health claims, on

average, consumers prefer a risk reduction claim and a disease prevention claim relative to a

function claim. This could occur if consumers believe a function claim is relatively general and

fairly weak (e.g. ―this product is good for your heart‖), and perhaps does not carry enough

information about the product‘s health benefit.

It was clear from the results that, on average, respondents preferred a risk reduction claim

relative to a disease prevention claim, even though the latter is a stronger claim. Perhaps

consumers perceive a disease prevention claim as a drug claim and are wary of this claim on a

food label. Also, Canadian consumers might be more familiar with a risk reduction claim on

certain functional foods in the actual market place (e.g. in Canada, products enriched with

calcium has been already labelled as ‗reducing the risk of osteoporosis‘) compared with a disease

prevention claim which is not permissible.

The results also show that respondents prefer to see verification of health claims, either by

the government or a third party. Consumers‘ preferences between government verification

(Health Canada) and a credible third party verification (Heart and Stroke Foundation) are similar,

and they are willing to pay price premiums for these verifications. These results suggest that

verification of health claims increases the credibility of health claims for Canadian consumers.

Put differently, Canadian consumers‘ trust in functional foods could be increased by the presence

91

of verifications of health claims on food labels. This has implications for how food

manufacturers might best market a new functional food. However, it will remain important for

government regulatory agencies to balance the needs of the food industry with the importance of

protecting consumers from misleading health claims or false or unsubstantiated verifications by

irresponsible third-party organizations or food companies.

As described in Chapter 4, the CL model in Table 5.2 provides a basic starting point for

analysis using more advanced discrete choice models, such as the Random Parameter Logit

model and the Latent Class Model. These models can explore different aspects of the consumer

choice framework.

5.2 CL Model with Interaction Effects

As discussed above, the model in Table 5.2 is a base model for all other estimation results

in this study. The estimation results in Table 5.4 extend the base model in Table 5.2 by adding in

interaction effects between the main attributes of the choice experiment. Thus, we can evaluate

the combined effect of health claims and source of verification, the combined effect of full and

partial labelling, etc.

92

Table 5.4: CL Model with Interaction Effects between the Main Attributes

Variable Coefficient t-ratio

Price -1.478*** -46.055

Function Claim .110 .700

Risk Reduction Claim .565*** 3.460

Disease Prevention Claim .458*** 2.843

Heart Symbol -.044 -.257

Government Verification .249** 2.128

Third Party Verification .431*** 3.941

Omega-3 .457*** 3.411

FC*Heart .346* 1.728

RRC*Heart .222 1.120

DPC*Heart .185 .941

RRC*GOV .223 1.389

RRC*TP -.251 -1.625

DPC*GOV -.012 -0.72

DPC*TP -.126 -.829

No Purchase -6.163*** -57.353

Log Likelihood Function = -5268.894; Pseudo-R2 = 0.251

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

As shown in equation (4.7) in Chapter 4, the design of the choice experiment in this study

allows an examination of the existence of both main effects and interaction effects for the

selected attributes. The model in Table 5.4 extends the base model in Table 5.2 by including the

interaction between full labelling and partial labelling, and also the interaction between disease-

related health claims and verification organizations. The values of the Pseudo-R2 are 0.25 in both

Table 5.4 and Table 5.2. Comparing the values of the Log Likelihood Functions in the two

models (LL= -5268.894 in Table 5.4 and LL= -5275.028 in Table 5.2), according to the

Likelihood Ratio (LR) test, the extended model in Table 5.4 has a borderline significance to

improve the goodness of fit of the model relative to the base model in Table 5.2.

93

Nevertheless, an interesting industry implication of labelling effects might be explored in

Table 5.4: whether the presence of a heart symbol strengthens or weakens health claims,

compared with the presence of either a function claim or a heart symbol, respectively. Among

the interaction variables in Table 5.4, FC*Heart represents the interaction effect between the

function claim and heart symbol. Only the coefficient of this variable (FC*Heart) is positive and

borderline significant at 10%, and all other coefficients for interaction variables are not

statistically significant. Louviere, Hensher and Swait (2000) suggested that an interaction effect

can be interpreted as an attribute having a complement/substitute relationship. Therefore, the

variables Function Claim and Heart Symbol are complements in this model. Why do consumers

value the interaction of function claim and heart symbol? It could be that the dual presence of

these attributes is a stronger signal or cue for a health benefit. However, this result might not be

robust. A post hoc analysis was conducted by taking out all the insignificant interaction variables

and estimating a new model with just the significant interaction effect (FC*Heart); the

coefficient of the FC*Heart becomes insignificant (see Appendix 1 for details). This may be

because, the coefficient of FC*Heart, as reported in Table 5.4, is only borderline significant at

the 10% level.

All other coefficients for the interaction variables in Table 5.4 are not significant,

indicating that consumers‘ utilities are not affected by these interaction effects. For example,

RRC*Heart has an insignificant coefficient, which means the combined effect of a risk reduction

claim and the heart symbol does not have an impact on consumers‘ utilities. In other words, risk

reduction claims may be strong enough on their own, adding a heart symbol to a risk reduction

claim does not add any significant value. Similar interpretations apply to the other insignificant

interaction variables. For example, the interaction variables between disease-related health

94

claims and verification organizations are also insignificant, while the individual coefficients for

health claims and verification organizations are all significant, indicating respondents do not

value any particular combination of health claims and verifications, but they respond positively

to each health claim or verification individually.

The model in Table 5.4 includes the interaction effects between the main attributes. The

base model could be further extended by including the interactions between the main attributes

and the covariates capturing consumers‘ attitudes or demographic information, as shown in the

following model.

5.3 CL Model with Interaction Effects for Covariates and Key Factors

Table 5.5(a) presents the results for the utility model specified in equation 4.9 of Chapter 4.

This model extends the base model (Table 5.2) by adding interaction effects between the main

variables and the exogenous covariates and the key factors. The total effect of each attribute

variable, when jointly considering the interaction effects and main effect, as reported in Table

5.5(a), is consistent with the base model reported in Table 5.2. The interaction effects between

the main variables and the exogenous covariates and the key factors could identify the sources of

consumer preference heterogeneity for certain attributes, as revealed in socio-demographic and

attitudinal differences. The exogenous covariates in the final model include consumers‘ income,

education, gender and whether they have heart disease. The three key factors are attitudes

towards functional foods, their health knowledge and their trust in health claims and nutrition

labels as described in Chapter 4.

The WTP estimates jointly consider the impacts of interaction variables and the original

variables for various attributes and can provide a richer and more comprehensive interpretation

95

of the corresponding coefficient estimates. Table 5.5(b) presents the WTP estimates for the CL

model with the covariates and the key factors from Table 5.5(a). The WTP measures are

calculated by the equation of marginal values (4.16) as introduced in Chapter 4.

Table 5.5 (a): CL Model with the Covariates and Key Factors

Variable Coefficient t-ratio Price -1.547*** -47.818

Function Claim 0.295*** 2.811

Risk Reduction Claim 0.761*** 6.367

Disease Prevention Claim 0.514*** 4.945

Heart Symbol 0.185*** 4.003

Government Verification 0.363*** 5.431

Third Party Verification 0.359*** 5.405

Omega-3 -0.083 -0.715

RRC*HeartDisease 0.537*** 2.730

DPC*HeartDisease 0.623*** 3.098

Omega-3*Income 0.005*** 6.050

RRC*Edu -0.136** -2.327

Omega-3*Gender 0.220*** 3.269

RRC*Attitude 0.139** 2.322

Omega-3*Attitude 0.655*** 17.347

Omega-3*Knowledge -0.113*** -3.521

RRC*Trust 0.224*** 4.198

GOV*Trust 0.301*** 5.469

No Purchase -6.374*** -58.469

Log Likelihood Function = -4956.556; Pseudo-R2 =0.295

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

96

Table 5.5 (b): WTP of CL Model with the Covariates and Key Factors

Variable WTP

($/2 litres)

t-ratio

Price - -

Function Claim

0.191*** 2.809

Risk Reduction Claim:

1) Without heart disease

2) With heart disease

3) With higher education (bachelors)

4) With lower education (high school or less)

5) With positive attitudes towards functional foods

6) With negative attitudes towards functional foods

7) With higher trust level

8) With lower trust level

9) Without heart disease, average education, average attitudes and average

trust level

10) Without heart disease, higher education, negative attitudes and lower

trust level

11) With heart disease, lower education, positive attitudes and higher trust

level

0.492***

0.839***

0.316***

0.492***

0.599***

0.324***

0.735***

0.288***

0.404***

-0.057

1.190***

6.400

6.036

3.986

6.400

6.951

2.941

7.695

3.141

5.913

-0.470

7.826

Disease Prevention Claim:

1) Without heart disease

2) With heart disease

0.332***

0.734***

4.945

5.322

Heart Symbol

0.120*** 3.998

Government Verification:

1) With higher level of trust

2) With lower level of trust

0.562***

-0.039

7.855

-0.580

Third Party Verification 0.232*** 5.421

Omega-3:

1) With higher income ($85,000)

2) With average income ($60,000)

3) With lower income ($35,000)

4) With positive attitudes towards functional foods

5) With negative attitudes towards functional foods

6) Male with average income, average attitudes and average knowledge

7) Female with average income, average attitudes and average knowledge

8) Male with lower income, negative attitudes and higher knowledge

9) Female with higher income, positive attitudes and lower knowledge

0.236***

0.150**

0.065

0.449***

-0.650***

0.150***

0.293***

-0.622***

0.969***

3.489

2.262

0.954

5.634

-7.651

2.262

4.793

-7.365

13.070

No Purchase

-4.120***

-69.273

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

97

The value of the Pseudo-R2 is 0.295 in Table 5.5(a) and 0.25 in Table 5.2. Comparing the

value of the Log Likelihood Function (LLF) in Table 5.5(a) (LL = -4956.556) with the value of

LLF in Table 5.2 (LL = -5275.028), according to the LR test, the extended model in Table 5.5(a)

significantly improves the model fit. Both the main effects and interaction effects are present in

Table 5.5(a). The estimation results are consistent with the base model as in Table 5.2. For the

main effects, all coefficients are significant at 1% with the expected signs, with the exception of

Omega-3, indicating that consumers prefer the presence of full labelling, partial labelling and

verification of health claims. The effect of consumers‘ preference for Omega-3 is captured by the

combined effects of both the main effect and the interaction effects for this attribute. Although

the coefficient is insignificant for the main effect of Omega-3, the coefficients of the effects of

the interactions between Omega-3 and other individual characteristics are most positive and

significant, such as Omega-3*Income, Omega-3*Gender and Omega-3*Attitude. Therefore,

taking together the combined effects for Omega-3 are still positive and significant, which implies

that in general consumers prefer Omega-3 to be present in milk products. For the estimation

results in Table 5.5(a), the interpretation focuses on the interaction effects. All coefficients of the

interaction effects are significant at 5% and with the expected signs.

In Table 5.5(b), the impacts of the attribute variables are illustrated in different scenarios

that consider demographic differences. Among those covariates and factors, there are continuous

and dummy variables. One practical method to calculate the marginal values is to use the sample

mean for continuous variables and separate into different scenarios according to the different

levels of dummy variables and calculate WTP at different levels of dummy variables. As

mentioned by Hu, Woods and Bastin (2009), since the goal is to find how different consumers

may value the product attributes differently, it would be interesting to see WTP at a high or low

98

level of a covariate in addition to the evaluation at mean values. In this study, when calculating

the WTP values in Table 5.5(b), two or three levels are chosen for each covariate or factor,

which are higher, average or lower levels. This study also considers scenarios with combinations

of different attribute levels.

The results show that consumers generally prefer full labelling over partial labelling.

Among full labelling, the risk reduction claim is more effective in increasing respondents‘

purchase intentions. Therefore, this model focuses on examining different individuals‘ responses

to the risk reduction claim by considering their demographic differences. Interactions between

Risk Reduction Claim and four other variables were examined in this model, including

respondents‘ pre-existing heart disease problems, their education levels, and factors 1 (attitude)

and factor 2 (trust).

The positive and significant coefficient for RRC*HeartDisease in Table 5.5(a), indicates

that respondents who self-reported that they are suffering from heart disease are sensitive to the

presence of a risk reduction claim on food labels. If a risk reduction claim is present on a milk

product it could significantly increase their purchase probabilities. In Table 5.5(b), according to

the corresponding WTP estimates for a risk reduction claim (the first and second scenarios), on

average, respondents without heart disease are willing to pay 49 cents and those with heart

disease are willing to pay 84 cents, for the presence of a risk reduction claim on a two-litre

carton of milk. The Wald test shows that the WTP difference between those two types of

respondents (with and without heart disease) is positive and significant, indicating that

respondents with heart disease are more likely to prefer the presence of a risk reduction claim,

compared with individuals who don‘t have heart disease. As one might expect, consumers with

heart disease are likely to pay more attention to health claims on food labels.

99

RRC*Edu in Table 5.5(a) is associated with a negative and significant coefficient,

indicating that respondents with higher levels of education are less likely to respond to a risk

reduction claim. In Table 5.5(b), according to the third and fourth scenarios for the risk reduction

claim, on average, respondents with higher levels of education (e.g. a bachelors degree) are

willing to pay 32 cents for a risk reduction claim on a two-litre carton of milk, compared to

individuals with lower levels of education (e.g. high school or less) who are willing to pay 49

cents for the presence of a risk reduction claim. The Wald test indicates that higher educated

respondents are less likely to prefer the risk reduction claim, relative to individuals with lower

education. One possible explanation might be that better educated consumers have greater levels

of scepticism towards risk reduction claims. This result is consistent with some marketing

literature in the area of scepticism and advertising. For example, DeLorme, Huh and Reid (2009)

conducted a study to investigate consumer advertising scepticism in the context of seeking

information sources for prescription drugs. They found that the overall level of direct-to-

consumer advertising scepticism was positively related to education, indicating that higher

educated consumers tend to have more scepticism towards advertising.

RRC*Attitude is the interaction between factor 1 (Attitude) and the risk reduction claim

(RRC). Factor 1 captures respondents‘ attitudes towards functional foods. The coefficient of

RRC*Attitude is positive and significant in Table 5.5(a), indicating that respondents with

positive attitudes towards functional foods are more likely to respond positively to risk reduction

claims. The WTP values of the risk reduction claim (the fifth and sixth scenarios) in Table 5.5(b),

show that respondents who have positive attitudes towards functional foods are willing to pay 60

cents for the presence of risk reduction claim on a two-litre carton of milk, whereas respondents

who have negative attitudes towards functional foods are willing to pay only 32 cents. According

100

to the Wald test, respondents with positive attitudes are more likely to prefer the presence of the

risk reduction claim on food labels, compared with respondents who have negative attitudes

towards functional foods. Notice that according to this result, even respondents with negative

attitudes towards functional foods could still respond positively to the presence of risk reduction

claims. This result implies that risk reduction claims are generally favoured by all respondents. It

could add more value to functional foods if this type of claim is permitted to be used on food

labels.

A similar interpretation applies to RRC*Trust. Factor 2 (Trust) captures respondents‘ trust

in health claims, nutrition labels and new food products. The coefficient of RRC*Trust is

positive and significant in Table 5.5(a). Respondents‘ WTP values for the presence of a risk

reduction claim on a two-litre carton of milk in Table 5.6(b) are 74 cents for those with higher

levels of trust and 29 cents for those with lower levels of trust. It implies that respondents who

tend to be more trust of health claims, nutrition labels and new food products are more likely to

pay more for the product if a risk reduction claim is present on food labels.

The last three scenarios of the WTP estimates for the risk reduction claim in Table 5.5(b),

show different combined levels of heart disease problems, education, attitudes towards

functional foods and trust in health claims. For example, in the ninth scenario, the WTP value is

calculated at the sample average level. Since the majority of respondents (more than 90%) of this

sample claim that they do not have heart disease, the combination of this scenario is for

respondents who do not have heart disease, with average education (college certificate or trade

diploma), average attitudes towards functional foods and average levels of trust on health claims.

On average, they would like to pay 40 cents for the risk reduction claim on a two-litre carton of

milk.

101

The lowest WTP for the risk reduction claim is shown in the tenth scenario in Table 5.5(b).

The WTP for the risk reduction claim is insignificant, among those respondents who do not have

heart disease, have higher education levels, with negative attitudes towards functional foods and

lower levels of trust in health claims. This result means that their WTP is not statistically

different from zero. The risk reduction claim seems not have any adverse effect and has the

potential to become a significant value-added attribute for functional food in Canada, as long as

this type of health claims is supported by sufficient scientific evidence.

In the eleventh scenario, it shows that the highest WTP for the risk reduction claims could

be obtained from those respondents who have heart disease, with lower education, with positive

attitudes towards functional foods and with higher levels of trust in health claims. They would

pay on average $1.19 for the presence of a risk reduction claim on a two-litre carton of milk.

This information might be particularly useful for functional food producers in terms of finding

target consumer groups and developing marketing strategies. However, this finding also calls

attention to the need for Canadian regulators to pay special attention to protecting this group of

consumers who might be particularly susceptible to misleading health claims.

The coefficient for DPC*HeartDisease, is also positive and significant in Table 5.5 (a).

The WTP values for the presence of a disease prevention claim on a two-litre carton of milk are

33 cents for respondents who do not have heart disease and 73 cents for respondents who have

heart disease in Table 5.5(b). According to the Wald test, the WTP difference between

individuals who have heart disease and who have not is positive and significant. Therefore, as

expected, the results clearly show that the presence of a disease prevention claim on food labels

is more likely to elicit a positive response from those consumers who suffer from heart disease.

102

Regarding the interaction effects with government verification, the coefficient of

GOV*Trust is positive and significant. Respondents with higher levels of trust are willing to pay

56 cents for government verification of health claims on a two-litre carton of milk, while the

WTP value is insignificant for individuals who have lower levels of trust. According to the Wald

test, the WTP difference between respondents with higher levels of trust and lower levels is

positive and significant. The result implies that respondents who tend to be more trusting of

health claims, nutrition labels and new food products are more likely to increase their purchase

intentions if a health claim is verified by a government organization.

Turning to the interaction effects for Omega-3, Omega-3*Income has a positive and

significant coefficient, indicating that respondents with higher incomes respond more positively

to Omega-3 milk products. The presence of Omega-3 on food labels could increase purchase

intentions of those consumers with higher incomes, while lower income consumers are less

sensitive to the presence of Omega-3 in milk products. As described in Table 5.1, the mean

household income across categories of the sample is Can $60,000. In Table 5.5(b), respondents‘

WTP values for the presence of Omega-3 on a two-litre carton of milk, are 24 cents by those

with higher incomes (e.g. Can $85,000), 15 cents by those with average incomes (Can $60,000,

the average incomes of the sample) and insignificant for lower income respondents (e.g. Can

$35,000). The Wald test shows that higher income individuals are more likely to prefer Omega-3

milk products relative to lower income consumers. If respondents‘ income is Can $35,000 or less,

the presence of Omega-3 on a milk product will not affect their purchase intention. This suggests

that higher income consumers are more likely to be able to afford the price premiums associated

with milk products with additional health benefits.

103

The coefficient of Omega-3*Attitude is positive and significant indicating that respondents

with positive attitudes towards functional foods are more likely to respond positively to Omega-3

milk products. As shown in Table 5.5(b), the WTP value for the presence of Omega-3 on a two-

litre carton of milk, are 45 cents for those respondents with positive attitudes towards functional

foods, and -65 cents for respondents with negative attitudes towards functional foods. Clearly,

respondents with positive attitudes towards functional foods tend to prefer Omega-3 milk, while

respondents with negative attitudes towards functional foods would need to be compensated by

65 cents to make them choose Omega-3 functional milk.

Omega-3*Knowledge is an interaction between Omega-3 and factor 3 (Health Knowledge),

which has a negative and significant coefficient in Table 5.5(a), indicating that respondents with

more health knowledge are less likely to respond to Omega-3 functional milk. The explanation

might be that respondents with more health knowledge have greater levels of scepticism towards

Omega-3 functional milk, or they might mainly consume Omega-3 ingredient from other food

sources rather than from Omega-3 milk. This result is consistent with some health

communication literature in the area of scepticism and knowledge. For example, Jallinoja and

Aro (2000) examined the impact of knowledge on respondent‘s attitude towards gene. Their

results showed that those respondents with the highest level of knowledge had more scepticism

than those with the lowest level of knowledge.

Omega-3*Gender has a positive and significant coefficient in Table 5.5(a), which implies

that female respondents are more sensitive to Omega-3 milk products relative to male

respondents. Notice that the WTP estimates for Omega-3 are not shown in separate scenarios of

just gender or health knowledge in Table 5.5(b). Although the coefficients of Omega-3*Gender

and Omega-3*Knowledge are all significant, WTP estimates indicate that these effects might not

104

be significant enough to translate into dollar values. Furthermore, the WTP estimates for Omega-

3 (the sixth and seventh scenarios) in Table 5.5(b), show that male respondents in the sample

with average incomes, average attitudes towards functional foods and average levels of health

knowledge are willing to pay 15 cents for the presence of Omega-3 on a two-litre carton of milk,

and female respondents with average income, attitudes and health knowledge are willing to pay

29 cents. The Wald test indicates that female respondents are more likely to prefer Omega-3 milk

relative to male respondents.

Two different groups of respondents with multiple demographic differences are presented in

the last two WTP scenarios, who might be willing to pay the lowest and highest amount of

money for the presence of Omega-3 on a two-litre carton of milk. It shows that male respondents

with lower incomes, negative attitudes towards functional foods and more health knowledge are

most likely to dislike the Omega-3 milk and would need to be compensated by 62 cents to make

them choose to consume Omega-3 milk. In comparison, female respondents with higher incomes,

positive attitudes towards functional foods and less health knowledge are most likely to prefer

Omega-3 milk and are willing to pay 97 cents for the Omega-3 attribute in a two-litre carton of

milk. This information could form the basis of market segmentation strategies for functional food

producers so as to focus on key target consumer groups.

In summary, the interaction effects in the CL model in Table 5.5(a) and Table 5.5(b) allow

for a more detailed examination of which types of respondents tended to respond positively or

negatively to the main attributes of functional foods examined in this study, including health

claims, verification organization, and the presence of Omega-3. Specifically, interactions

between a risk reduction claim and four other variables were explored, including heart disease

problems, education levels, factor 1 (attitude) and factor 2 (trust). The results indicate that the

105

effect of having a risk reduction claim on food labels is intensified among respondents self

reporting as suffering from heart disease, possessing relatively lower education levels, having

positive attitudes towards functional foods, and who trust health claims and nutrition labels.

Furthermore, interactions between the presence of Omega-3 and four other variables were also

explored: income, gender, factor 1 (attitude) and factor 3 (knowledge). The presences of Omega-

3 in functional milk products is more likely to engender a positive response from consumers who

are female, have higher incomes, have positive attitudes towards functional foods and consider

themselves less knowledgeable about health. Consumers who tend to trust health claims and

nutrition labels are more likely to respond positively to the government verification of health

claims. These findings provide a starting point for the food industry or government regulatory

agencies to begin to understand consumers‘ heterogeneous responses to health claims and the

verifications of those health claims.

5.4 Base Model: the Random Parameter Logit Model and WTP

Table 5.6 presents the estimation results for the base model using the Random Parameter

Logit (RPL) model, which is used to capture consumers‘ preference heterogeneity. Compared

with the results of the CL model in Table 5.2, according to the LR test the goodness of fit of this

model has a significant improvement in terms of the values of the Log Likelihood Function (-

5275.028 in Table 5.2 v.s. -3845.335 in Table 5.6) and the Pseudo-R2 (0.25 in Table 5.2 v.s.

0.531 in Table 5.6). As explained in Chapter 4, the values of the estimated parameters are fixed

in the Conditional Logit model. In the RPL model, however, the estimated parameters are

continuous and follow a certain distribution, and the density of the parameters can be specified to

be normal or some other distribution (Train, 2003). There are two sets of estimated parameters in

a RPL model. One set includes the estimated random/fixed parameters of the examined attributes,

106

and the other set represents the standard deviations of the random parameters as shown in Table

5.6.

Table 5.6: Base Model: RPL and WTP Estimates

Variable Coefficient t-ratio WTP

($/2 litres)

t-ratio

Mean value of Random parameters in utility function Omega-3 1.300*** 6.539 0.524*** 6.658

Risk Reduction Claim 1.043*** 7.106 0.420*** 7.200 Disease Prevention Claim 1.034*** 6.833 0.416*** 6.913

Heart Symbol 0.313*** 4.653 0.126*** 4.685 Government Verification 0.903*** 8.225 0.364*** 8.359 Third Party Verification 0.547*** 5.428 0.220*** 5.476

Mean value of Fixed parameters in utility function

Price -2.483*** -37.198 - - Function Claim 0.316** 2.162 0.127** 2.161

No Purchase -9.214*** -44.364 -3.711*** -93.638

Standard deviations of random parameters Sd-Omega-3 4.046*** 22.490 - -

Sd-Risk Reduction Claim 0.641*** 4.067 - - Sd-Disease Prevention Claim 1.258*** 9.754 - -

Sd-Heart Symbol 0.525*** 4.374 - - Sd-Government Verification 1.104*** 7.792 - - Sd-Third Party Verification 0.663*** 3.453 - -

Log Likelihood Function = -3845.335; Pseudo-R2 = 0.531

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

The estimation results in Table 5.6 are generally consistent with the results in Table 5.2 for

the examined attributes. All random and fixed parameters are highly significant and with the

expected signs. Consumers prefer all three types of health claims relative to no health claim

(base level). They also prefer a heart symbol (partial labelling) to be present on food labels

relative to no symbol. The respondents also prefer the verifications of health claims by

government or third party compared with no verification (base level). They respond positively to

107

milk enriched with Omega-3. Not buying any milk product or paying a higher price decreases

their utilities.

However, some differences exist in terms of WTP estimations in the two models presented

in Tables 5.2 and 5.6. First, when the base model is estimated by the Random Parameter Logit

model in Table 5.6, consumer‘s WTP for government verification is 36 cents for a two-litre

carton milk, which is significantly higher than their WTP for third party verification at 22 cents,

indicating that government verification has a bigger impact on consumers‘ utilities than third

party verification when heterogeneity of preferences is taken into account. In contrast, the model

presented in Table 5.2 was unable to detect a significant difference between consumers‘ WTP for

these two verification organizations. Second, the WTP values for a risk reduction claim and a

disease prevention claim are close to each other, at about 42 cents for a two-litre carton of milk

in the RPL model (Table 5.6). However, in the CL model (Table 5.2), consumers‘ WTP for a risk

reduction claim is significantly higher than for a disease prevention claim. Third, consumers‘

WTP for Omega-3 in Table 5.6 is higher than in Table 5.2. These differences are the results of

using different estimation methods. The fixed coefficients in the CL model can only estimate

average values which assume that respondents‘ preferences are homogenous, while the random

parameters in the RPL model allow for heterogeneity of preferences which tends to be more

realistic. The goodness of fit of the RPL model is also better than the CL model.

The estimated standard deviations of the random parameters are all highly significant,

which further indicates that it is necessary to include a mixed structure for the selected

parameters in the CL model to improve the goodness of fit of the model. The significant standard

deviations of the random parameters indicate that strong variations exist in consumers‘

preferences for whether the milk product is enriched with Omega-3, whether there is a risk

108

reduction claim or a disease prevention claim on food labels, whether a heart symbol is present,

and whether there is government or third party verification of the health claims.

The random parameters of the RPL model further identify the distribution of individual

taste preferences in the population. For example, for a normal distributed parameter, which has a

positive (negative) mean estimate, the share of respondents who have a positive (negative) view

of that attribute can be calculated. Following Train (2003) and Hu, Veeman and Adamowicz

(2005), for a random parameter β ~ N(b, σ2), the probability of β < 0, equals Φ[(0-b)/σ]. Φ (β | b,

σ2) is the normal density of the standard normal distribution. The probability of β > 0, equals 1-

Φ[(0-b)/σ]. The interpretation of the probabilities of β < 0 (> 0) is the percentage/share of

respondents who have a negative (positive) view of an attribute. The RPL estimation results in

Table 5.6 were obtained by assuming a normal distribution of the random parameters for the

variables Omega-3, Risk Reduction Claim, Disease Prevention Claim, Heart Symbol,

Government Verification and Third Party Verification. Table 5.7 reports respondents‘

heterogeneous preferences for the attributes with random parameters.

Table 5.7: Respondents‟ Heterogeneous Preferences

for the Attributes with Random Parameters

Variable Positive Percentage Negative Percentage

Omega-3 62.55% 37.45%

Risk Reduction Claim 94.84% 5.16%

Disease Prevention Claim 79.39% 20.61%

Heart Symbol 72.57% 27.43%

Government Verification 79.39% 20.61%

Third Party Verification 79.67% 20.33%

109

In total, 62.55% of the respondents have positive views (37.45% with negative views) of the

Omega-3 attribute, indicating that about two thirds consumers prefer the milk products

containing Omega-3. The majority (94.84%) of respondents value a risk reduction claim

positively relative to no health claim. There are 79.39% of respondents who value a disease

prevention claim positively relative to no health claim. The heart symbol attribute is positively

viewed by 72.57% of respondents. There are 79.39% and 79.67% of respondents who have

positive values for government verification and third party verification, respectively, indicating

that most consumers prefer government or third party verification relative to no verification.

Clearly the respondents have different percentages of positive views (ranging from 62.55% to

94.84%) for the attributes listed in Table 5.7, indicating different degree of heterogeneity in

consumers‘ responses to these attributes.

5.5 Base Model: LCM Estimates

Recognizing the limitations of the CL model that were discussed earlier, the Random

Parameter Logit and Latent Class Models were used to incorporate heterogeneity in consumers‘

choices with a significant improvement in model fit compared with the CL results. Table 5.8

presents the estimation results of the LCM and WTP for the base (main effects) model. As

discussed in Chapter 4, the CL model assumes that all consumers‘ preferences are homogeneous.

While the LCM assumes that respondents can be intrinsically grouped into a number of latent

classes where, in each class, consumers‘ preferences are still assumed to be homogenous but

heterogeneous across classes (Boxall and Adamowicz, 2002). To identify the number of latent

classes the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) are

employed. Both AIC and BIC are minimized when the number of latent classes is equal to four

for this model, as presented in Table 5.8. The value of the Pseudo-R2 is 0.61 in Table 5.8 and

110

0.25 for the CL model in Table 5.2. The value of the Log Likelihood Function (LLF) in this

model is -3204.068, compared with the value of LLF in Table 5.2 (LL = -5275.028). Therefore,

according to the LR test, the model fitness is significantly improved in the LCM compared to the

CL model in Table 5.2.

Table 5.8: Base Model: LCM Estimation Results

Class 1:

Price Sensitive

Consumers

Class 2:

Functional Milk

Believers

Class 3:

Verification Seekers

Class 4:

Health Claims

Rejecters

Variables Coefficients WTP Coefficients WTP Coefficients WTP Coefficients WTP

Price -2.694***

(-34.528)

- -2.838***

(-28.169)

- -.433***

(-8.839)

- -2.343***

(-12.728)

-

Function

Claim

.005

(.024)

.002

(.024)

.937***

(4.776)

.33***

(4.718)

1.074***

(4.324)

2.477***

(3.748)

-1.183**

(-2.304)

-.505**

(-.305)

Risk

Reduction

Claim

.097

(.430)

.036

(.429)

1.122***

(5.836)

.395***

(5.83)

2.607***

(10.3)

6.014***

(6.649)

-.861

(-1.608)

-.368

(-1.606)

Disease

Prevention

Claim

-.031

(-.140)

-.011

(-.140)

1.467***

(7.125)

.517***

(7.218)

2.498***

(10.35)

5.765***

(6.804)

-.386

(-.773)

-.165

(-.769)

Heart

Symbol

-.025

(-.202)

-.009

(-.203)

.094

(1.084)

.033

(1.084)

.429***

(7.144)

.989***

(5.805)

.205

(.729)

.088

(.735)

Government

Verification

-.023

(-.143)

-.009

(-.143)

.236

(1.383)

.083

(1.410)

1.318***

(12.527)

3.042***

(7.781)

.777**

(2.299)

.331**

(2.373)

Third Party

Verification

.108

(.692)

.040

(.692)

.384**

(2.471)

.135**

(2.511)

.829***

(7.858)

1.913***

(5.759)

.689**

(2.042)

.294**

(2.064)

Omega-3 -.016

(-.080)

-.006

(-.080)

5.05***

(28.519)

1.78***

(29.463)

1.603***

(4.856)

3.698***

(4.775)

1.247***

(2.828)

.532***

(2.858)

No Purchase -11.35***

(-39.642)

-4.21***

(-47.136)

-8.57***

(-22.354)

-3.02***

(-26.97)

-.946***

(-3.094)

-2.18***

(-3.359)

-4.809***

(-11.574)

-2.05***

(-48.87)

Log Likelihood Function = -3204.068. Pseudo-R2 = 0.61; T-ratios in brackets.

Estimated Latent Classes Probabilities (Number of Latent Classes = 4)

Coefficient

(t-ratio)

Coefficient

(t-ratio)

Probability of Class 1 .550***

(15.883)

Probability of Class 3 .148***

(10.589)

Probability of Class 2 .256***

(21.595)

Probability of Class 4 .046***

(3.345)

In this LCM, there exist four different preference groups with the estimated latent classes‘

probabilities of 55%, 25%, 15% and 5%. As Nilsson, Foster and Lusk (2006) explain these are

111

the probabilities of a randomly chosen respondent belonging to the first, second, third or fourth

class, respectively.

In the first latent class in Table 5.8, only the coefficients for Price and No Purchase are

statistically significant and with the expected signs. Other coefficients are not significant. These

results indicate that consumers within this class do not value Omega-3 milk or any of the

labelling attributes (including both full labelling and partial labelling) or verifications. These

consumers appear to be indifferent to functional milk with Omega-3 or conventional milk. Price

is the only determining factor when they make milk purchase decisions. Therefore, this class is

referred to as ―Price Sensitive Consumers‖. Compared with other classes, the WTP for No

Purchase is the highest amount for this group, indicating that respondents‘ utilities in this class

suffer the largest discount from not buying a milk product. There is a 55% probability that a

randomly chosen individual belongs to this class, indicating that a slight majority of respondents

(at least 55% of the sample population) still have no clear preference between conventional milk

and Omega-3 enriched milk.

In the second latent class almost all coefficients are statistically significant and with the

expected signs, with the exception of the variables Heart and Government Verification. There is

a 25% probability that a randomly chosen individual belongs to this class. Consumers in this

class prefer all three types of full health claims over the absence of a health claim, but they do

not respond to partial labelling. The WTP values are 33 cents, 40 cents and 52 cents for the

function claim, the risk reduction claim and the disease prevention claim, respectively, indicating

that these consumers have a stronger preference for a disease prevention claim. Consumers in

this class prefer Omega-3 enriched milk and are willing to pay a fairly high amount ($1.78) for

this attribute. This class is named ―Functional Milk Believers‖. Consumers in this class also

112

prefer verification by a third party and are willing to pay 14 cents for this attribute, but they do

not value government verification.

In the third latent class all of the coefficients are highly significant and with the expected

signs. Consumers in this class respond positively to full labelling, partial labelling, verification

organizations and the presence of Omega-3. The WTP estimated for these attributes are very

high. For example, these consumers are willing to pay price premiums as much as $2.48, $6.01,

$5.77 and $0.99 for a function claim, a risk reduction claim, a disease prevention claim and a

heart symbol, respectively, indicating that those consumers have a stronger preference for a risk

reduction claim. These consumers are willing to pay $3.04, $1.91 and $3.70 for the government

verification, third party verification and the presence of Omega-3. Unlike consumers in the

second class, consumers in this class have a strong preference for government verification as

indicated by its highly positive and significant WTP value. Within this class, consumers are more

likely to prefer government verification of health claims relative to third party verification. These

consumers‘ WTP for the presence of Omega-3 are also the highest amount within the four

classes. Compared with other classes, respondents within this group most concern about and are

willing to pay the highest amount for the verification of health claims. Therefore, this class is

referred to as ―Verification Seekers‖. There is a 15% probability that a randomly chosen

individual belongs to this class. The WTP values for consumers in this class seem unrealistically

high and may reflect one of the shortcomings of the hypothetical nature of the choice experiment.

These WTP estimates are perhaps better interpreted as indications of the relative strength of

preferences rather than absolute values. As shown in the next section, adding covariates into the

estimation of the LCM appears to be a good way to derive more reasonable WTP estimates.

113

In the fourth latent class, the coefficients for risk reduction and disease prevention claims

are negative but not significant. The effect of Function Claim is significant, but negative. The

coefficient for Heart Symbol is positive but not significant. It appears that consumers in this

group do not believe in either full labelling or partial labelling. They especially dislike the

function claim. Therefore, this class is referred to as ―Health Claims Rejecters‖. These

consumers prefer the presence of Omega-3, but they would like to see verifications, either by the

government or a third party. One possible interpretation of the behaviours of consumers in this

group is that they might not value any type of health claim without verification; they only

respond positively to verification. There is only a 5% probability that a randomly chosen

individual belongs to this class.

In summary, according to the estimation results in Table 5.8, consumers‘ preferences are

heterogeneous for full labelling, partial labelling, verification organizations and the presence of

Omega-3 across four latent classes. For example, for the price sensitive consumers in the first

class, they do not value any labelling or the Omega-3 attribute. Price is the only determining

factor when they make milk purchase decisions. Data indicate that at least 55% of the population

belongs to this group who have no particular preference between conventional milk and Omega-

3 milk. Therefore, reducing price is likely to be the most efficient strategy to make this group of

consumers purchase Omega-3 milk. Comparing consumers‘ preferences for full labelling with

partial labelling, three explicit health claims are preferred by consumers in class 2 and 3. The use

of a Heart Symbol (partial labelling), is only preferred by the third class. Therefore, according to

the estimation of this LCM model, full labelling is an effective communication tool to about 40%

of the population, whereas, partial labelling can only effectively communicate with 15% of the

population.

114

5.6 LCM Model with Interaction Effects of Covariates and Key Factors

Table 5.9(a) represents the LCM estimation results for an extended model with the

interaction effects between some of the main variables and the exogenous covariates and the key

factors. The AIC and BIC criteria are employed to identify the number of latent classes in this

LCM. Both AIC and BIC are minimized when the number of latent classes is equal to four. The

value of the Pseudo-R2 is 0.622 in this model, compared with 0.61 in the model presented

previously in table 5.8. The value of the Log Likelihood Function (LLF) in this model is -

3100.954, compared with the value of LLF (LL = -3204.068) in table 5.8. According to the LR

test, this model has a significant improvement in the goodness of fit over the LCM base model in

table 5.8. Table 5.9 (b) presents the WTP estimates for the total effect of each attribute according

to the estimation results in table 5.9 (a). Given the hypothetical nature of the choice experiment,

the WTP estimates are most usefully interpreted as relative measures of the strength of

preferences. The WTP estimates in table 5.9 (b) allows a comparison of the impacts of various

attributes both within and cross different latent classes.

The interaction effects contained in Table 5.9 (a) work as explanatory variables to further

identify consumer heterogeneity. For the main attributes reported in table 5.9 (a) and the WTP

estimates for the total effect of each attribute reported in Table 5.9 (b), consumers‘ preferences

for the product attributes in each latent class are consistent with the estimation results in Table

5.8. Accordingly, the names of those latent classes in Table 5.8 are still suitable for this model,

although of course the class probabilities are different.

115

Table 5.9 (a): LCM with Interaction Effects for the Covariates and Key Factors

Class 1:

Price Sensitive

Consumers

Class 2:

Functional Milk

Believers

Class 3:

Verification

Seekers

Class 4:

Health Claims

Rejecters

Variable

Coefficient

(t-ratio)

Coefficient

(t-ratio)

Coefficient

(t-ratio)

Coefficients

(t-ratio)

Price -3.071*** (-29.029)

-3.700*** (-21.138)

-0.880*** (-22.723)

-1.981*** (-15.583)

Function Claim 0.075 (0.311)

0.510* (1.709)

0.470*** (2.849)

-0.086 (-0.219)

Risk Reduction Claim 0.030 (0.102)

1.480*** (4.032)

1.734*** (8.952)

-0.146 (-0.319)

Disease Prevention Claim -0.022 (-0.093)

1.198*** (3.720)

1.653*** (10.258)

0.319 (0.804)

Government Verification -0.099 (-0.522)

0.536** (2.062)

0.870*** (9.047)

0.873*** (3.885)

Third Party Verification 0.067 (0.394)

1.061*** (4.181)

0.666*** (7.433)

0.373 (1.402)

Heart Symbol -0.014 (-0.097)

0.040 (0.309)

0.273*** (5.424)

0.647*** (3.671)

Omega-3 -0.214 (-0.766)

2.350*** (5.747)

1.688*** (5.144)

0.617 (1.641)

RRC*Attitude 0.104 (0.608)

-0.468 (-1.638)

0.298*** (4.219)

0.472* (1.658)

Omega-3*Attitude 0.217** (2.513)

14.462*** (16.518)

0.152 (1.174)

1.470*** (11.317)

Omega-3*Knowledge -0.144** (-2.015)

-2.018*** (-10.380)

-0.344** (-2.421)

-0.448*** (-4.645)

RRC*Trust -0.011 (-0.070)

-0.111 (-0.578)

0.295*** (3.678)

-0.240 (-1.078)

GOV*Trust 0.090 (0.609)

0.252 (1.173)

0.267*** (4.001)

0.780*** (4.735)

RRC*HeartDisease 0.484 (0.871)

0.301 (0.421)

-0.669** (-2.431)

14.113 (0.545)

DPC*HeartDisease -0.034 (-0.064)

-0.014 (-0.016)

-0.130 (-0.492)

12.363 (0.478)

Omega-3*Income 0.00499** (2.489)

0.00735 (1.591)

0.00548 (1.387)

-0.0104*** (-3.909)

RRC*Edu -0.004 (-0.023)

-0.240 (-1.113)

-0.276*** (-2.993)

0.149 (0.620)

Omega-3*Gender 0.473*** (2.779)

7.845*** (14.880)

0.861*** (2.952)

1.791*** (9.510)

No Purchase -12.518*** (-34.60)

-14.042*** (-22.659)

-4.697*** (-8.840)

-3.780*** (-12.980)

Log Likelihood Function = -3100.954; Pseudo-R2 = 0.622;

Estimated Latent Classes Probabilities (Number of Latent Classes = 4)

Coefficient

(t-ratio)

Coefficient

(t-ratio)

Probability of Class 1 0.491***

(34.776)

Probability of Class 3 0.216***

(13.179)

Probability of Class 2 0.226***

(13.977)

Probability of Class 4 0.067***

(3.787)

116

Table 5.9 (b): WTP of LCM with Interaction Effects for the Covariates and Key Factors

Class 1:

Price

Sensitive

Consumers

Class 2:

Functional

Milk

Believers

Class 3:

Verification

Seekers

Class 4:

Health

Claims

Rejecters

Variable

WTP

($/2 litres)

WTP

($/2 litres)

WTP

($/2 litres)

WTP

($/2 litres)

Price - - - -

Function Claim 0.0243

(0.311)

0.138*

(1.704)

0.534***

(2.822)

-0.043

(-0.219)

Risk Reduction Claim:

1) With heart disease, average education level,

average attitudes and trust level

2) Without heart disease, average education level,

average attitudes and trust level

0.166

(0.896)

0.009

(0.106)

0.417**

(2.119)

0.335***

(4.073)

0.897***

(2.604)

1.657***

(8.451)

7.125

(0.545)

0.001

(0.007)

Disease prevention Claim:

1) With heart disease

2) Without heart disease

-0.018

(-0.101)

-0.007

(-0.093)

0.320

(1.361)

0.324***

(3.748)

1.730***

(5.206)

1.878***

(10.243)

6.402

(0.490)

0.161

(0.806)

Heart Symbol -0.004

(-0.097)

0.011

(0.309)

0.310***

(5.472)

0.326***

(3.961)

Government Verification

with average trust level

-0.032

(-0.525)

0.145**

(2.132)

0.989***

(9.584)

0.441***

(3.973)

Third Party Verification 0.022

(0.394)

0.287***

(4.456)

0.757***

(7.254)

0.188

(1.404)

Omega-3:

1) Female with average income, average attitudes

and health knowledge

2) Male with average income, average attitudes

and health knowledge

0.182***

(2.580)

0.028

(0.347)

2.875***

(22.458)

0.754***

(9.090)

3.268***

(10.012)

2.291***

(9.031)

0.902***

(5.322)

-0.002

(-0.012)

No Purchase -4.077***

(-43.398)

-3.795***

(-81.657)

-5.335***

(-8.808)

-1.908***

(-35.472)

Estimated Latent Classes Probabilities (Number of Latent Classes = 4); T-ratios in brackets.

Coefficient

(t-ratio)

Coefficient

(t-ratio)

Probability of Class 1 0.491***

(34.776)

Probability of Class 3 0.216***

(13.179)

Probability of Class 2 0.226***

(13.977)

Probability of Class 4 0.067***

(3.787)

117

In the first latent class (Price Sensitive Consumers) in Table 5.9(a) and in Table 5.9(b),

there is a 49% probability of a respondent falling into this group. In terms of the main effects,

only Price and NoPurchase have significant coefficients with the expected signs. The

respondents in class 1 are relatively uninterested in Omega-3 milk and do not value any of the

full labelling, partial labelling, verification organizations or Omega-3. They are more likely to be

conventional milk consumers. Omega-3*Attitude, Omega-3*Income and Omega-3*Gender are

interactions between the presence of Omega-3 and three other variables. They are the only three

positive and significant interaction variables in class 1, indicating that those respondents with

positive attitudes towards functional foods, with higher incomes or being female are still more

likely to positively respond to the presence of Omega-3 in milk products, although on average,

the respondents in this class are price sensitive. Omega-3*Knowledge is an interaction between

Omega-3 and the health knowledge factor (factor 3). Notice that, as shown in Table 5.9(a), the

coefficients of Omega-3*Knowledge are negative and significant across four classes, indicating

that respondents with higher health knowledge in this sample are more likely to be scepticism

toward Omega-3 functional milk, compared with those with lower health knowledge.

Furthermore, in Table 5.9(b), respondents‘ WTP estimates for the main attributes are all

insignificant in class 1, with the exception of one group of respondents‘ WTP for Omega-3. This

group of respondents are female with average incomes, average attitudes towards functional

foods and average health knowledge. On average, they are willing to pay 18 cents for the

Omega-3 attribute on a two-litre carton of milk. Compared with the respondents with similar

social-demographic and attitudinal characteristics in other classes, the WTP of this group for

Omega-3 is the smallest amount relative to the other classes.

118

Class 2, labelled as Functional Milk Believers, consists of a group of respondents who

moderately prefer Omega-3 milk, as shown by the positive and significant coefficients and WTP

estimates for the Omega-3 attribute. Those respondents prefer full labelling relative to partial

labelling, indicated by the positive and significant WTP estimates for the three explicit health

claims, and by the insignificant coefficient and WTP estimate for partial labelling. Table 5.9(b)

shows that the WTP for a disease prevention claim is similar with their WTP for a risk reduction

claim for the respondents who do not have heart disease. For those respondents who have heart

disease, their WTP for a risk reduction claim is higher than for a disease prevention claim. This

group of respondents value the verifications of health claims both by the government and a third

party. Their WTP is relatively higher for third party verification, while respondents in other

classes are willing to pay more for the government verification than they are for verification by a

third party. There is a 22.6% probability that a respondent would belong to this class.

For the interaction effects in class 2, the coefficient of Omega-3*Attitude is positive and

significant, indicating respondents with positive attitudes towards functional food in this class

tend to respond more positively to the presence of Omega-3 in milk products. The negative and

significant coefficient of Omega-3*Knowledge captures the effect that respondents with higher

levels of self-reported health knowledge tend to discount the presence of Omega-3 in milk

relative to respondents with lower health knowledge. Omega-3*Gender has positive and

significant coefficient, which implies that female respondents exhibit stronger preferences for

Omega-3 milk.

In the third latent class of Tables 5.9(a) and 5.9(b), respondents exhibit relatively strong

preferences for Omega-3, the presence of a health claim, and the verifications of health claims.

The WTP values for those attributes are fairly high in this class as reported in Table 5.9(b). In

119

fact, they are the highest levels among the four classes. Particularly, these respondents exhibit

the strongest WTP for verification of health claims. Therefore, this class is again named as the

‗Verification Seekers‘. There is a 21.6% probability of a randomly chosen respondent falling into

this class. As revealed by the higher WTP estimates for full labelling and partial labelling, these

respondents‘ preferences for the labelling effects are the highest across the four classes. Both

government and third party verifications are valued by respondents in this group, with a higher

WTP for government verification. Unlike class 2, respondents in this class also value the

presence of a red heart symbol on food labels.

With respect to the interaction effects, respondents who self-declare higher levels of trust

in health claims and nutrition labels (factor 2) tend to respond more positively to the government

verification and risk reduction claims. Respondents in this group with higher levels of education

also tend to discount risk reduction claims. Respondents in this segment with positive attitudes

towards functional foods (factor 1) are more likely to prefer risk reduction claims. Female

respondents in this group tend to respond positively to the presence of Omega-3 in milk products.

Finally, class 4 is a relatively small segment with a 6.7% probability that a randomly

chosen individual belongs to this class. Respondents in this class are ‗Health Claim Rejecters‘,

since they do not value any type of full health claims. However, they prefer partial labelling,

indicated by the positive and significant coefficient and WTP. Females are more likely to prefer

Omega-3 milk relative to male respondents in this class, when these respondents are evaluated at

average income, average attitudes towards functional foods and average health knowledge levels.

This class values government verification more than third party or no verification. However,

respondents who self-declared higher levels of health and nutrition knowledge (factor 3) and

respondents with higher income levels, tend to discount the presence of Omega-3 in milk

120

products, which may reveal some scepticism among those respondents with respect to the health

effects associated with Omega-3 milk.

The LCM presented in Tables 5.9(a) and 5.9(b) incorporates covariates and key factors

through interaction terms to capture the heterogeneity within each latent class. The following

LCM with membership indicators extends the LCM base model to identify the source of

heterogeneity across the latent classes.

5.7 The LCM Model with Class Membership Indicators

Preference heterogeneity is critical to understanding consumer demand for functional foods.

It is difficult to examine the heterogeneity in a random utility model because an individual‘s

characteristics are invariant among the choice sets in the choice experiment. One way to solve

this problem is by interacting individual characteristics with the main attributes in the choice

experiment, as shown in Table 5.5(a) and 5.9(a). Adamowicz et al. (1997) adopted this method to

identify preference heterogeneity. The limitations of using the interaction variables method,

however, are that a priori selection of key individual characteristics and attitudes is necessary,

and only a limited selection of individual specific variables can be involved (Boxall and

Adamowicz, 2002). Another method by which to identify preference heterogeneity is to use the

Random Parameter Logit model to allow the estimated parameters to vary randomly across

individuals. The limitation of this approach is that it is not well-suited for explaining the sources

of heterogeneity and those sources might closely relate to the characteristics of individual

consumers (Boxall and Adamowicz, 2002).

A promising way to solve these problems is to extend the Latent Class Model to consider

the observed constituent variables of the class membership (McCutcheon 1987; Boxall and

121

Adamowicz, 2002). The LCM with class membership indicators offers a better way to

incorporate and explain the sources of preference heterogeneity. Notice that the method of

interacting individuals‘ characteristics with main attributes in the LCM focuses on capturing

heterogeneity within each latent class, such as that shown in Table 5.9(a); the LCM with class

membership indicators focuses on explaining the sources of heterogeneity across the latent

classes, which provides added insights as to the sources of heterogeneity. Table 5.10 presents the

results of the LCM with class membership indicators.

122

Table 5.10: LCM with the Class Membership Indicators

Class 1:

Price Sensitive

Consumers

Class 2:

Functional Milk

Believers

Class 3:

Verification Seekers

Class 4:

Health Claims

Rejecters Variable

Coefficient WTP Coefficient WTP Coefficient WTP Coefficient WTP

Price -2.52*** (-35.121)

-

-2.73*** (-29.222)

- -0.394*** (-7.601)

- -1.56*** (-10.284)

-

Function

Claim -0.035

(-0.168) -0.014

(-0.168) 0.891*** (4.625)

0.326*** (4.582)

1.142*** (4.351)

2.895*** (3.616)

-1.43*** (-2.843)

-0.916*** (-2.763)

Risk

Reduction

Claim

0.082 (0.395)

0.032 (0.395)

1.134*** (5.956)

0.415*** (5.969)

2.690*** (10.067)

6.821*** (6.005)

-0.377 (-0.760)

-0.241 (-0.753)

Disease

Prevention

Claim

-0.023 (-0.116)

-0.009 (-0.116)

1.435*** (7.142)

0.525*** (7.257)

2.573*** (10.049)

6.524*** (6.139)

-0.158 (-0.327)

-0.101 (-0.326)

Government

Verification 0.026

(0.178) 0.010

(0.178) 0.195

(1.195) 0.072

(1.213) 1.275*** (11.258)

3.232*** (6.809)

1.825*** (5.793)

1.167*** (5.970)

Third Party

Verification 0.164

(1.151) 0.065

(1.150) 0.375** (2.517)

0.137** (2.553)

0.890*** (7.892)

2.256*** (5.298)

0.842** (2.295)

0.539** (2.320)

Heart

Symbol -0.024

(-0.210) -0.009

(-0.210) 0.090

(1.065) 0.033

(1.066) 0.430*** (6.827)

1.091*** (5.268)

0.636*** (2.713)

0.407*** (2.835)

Omega-3 0.011 (0.061)

0.004 (0.061)

4.864*** (28.585)

1.781*** (28.854)

2.409*** (4.889)

6.107*** (4.554)

0.875** (2.145)

0.560** (2.110)

No Purchase -10.2*** (-42.922)

-4.06*** (-56.041)

-8.42*** (-22.206)

-3.081*** (-26.378)

-2.106** (-2.563)

-5.340** (-2.502)

-2.37*** (-6.530)

-1.514*** (-13.988)

Class 1:

Price Sensitive

Consumers

Class 2:

Functional Milk

Believers

Class 3:

Verification Seekers

Class 4:

Health Claims

Rejecters

Constant 2.005*** (7.495)

0.991*** (3.352)

0 -0.941 (-1.607)

Income -0.0095*** (-2.725)

-0.0042 (-1.095)

0 -0.0048 (-0.555)

Education 0.230 (1.516)

0.130 (0.774)

0 -0.159 (-0.425)

Heart

Disease -0.790* (-1.826)

-0.627 (-1.359)

0 -0.944 (-0.821)

Attitude -0.770*** (-5.350)

-0.115 (-0.739)

0 -0.801*** (-2.714)

Trust -0.567*** (-4.297)

-0.426*** (-2.985)

0 -0.358 (-1.286)

Log Likelihood Function = -3174.479. Pseudo-R2 = 0.613; T-ratios in brackets.

Estimated Latent Classes Probabilities (Number of Latent Classes = 4)

Coefficient Coefficient

Probability of Class 1

0.569 Probability of Class 3 0.137

Probability of Class 2

0.266 Probability of Class 4 0.028

123

Rather than including the exogenous covariates and factors as interaction effects, Table

5.10 presents the base model with these variables used to explain class membership. Those

covariates such as income, education, health disease and key factors from the Factor Analysis are

used to help explain class membership. The AIC and BIC criteria are used to identify the number

of latent classes in this LCM. Again, four latent classes are identified. The estimation results in

Table 5.10 have a Log Likelihood Function of -3174.479 with a Pseudo-R2 of 0.613, indicating

the model is a relatively good fit. The estimated class probabilities for the four latent classes are

56.9%, 26.6%, 13.7% and 2.8%, respectively. Two sub-tables are included in Table 5.10, the top

half is the estimated coefficients and WTP values for the four segments, while lower section

contains the estimated parameters for the class membership indicators. The estimated

coefficients and WTP values for the four latent classes in Table 5.10 are consistent with the

estimation results of the LCM for the base model in Table 5.8, which means that the distributions

of consumers‘ preference heterogeneity are very similar in both models. Therefore, the adopted

group names for Tables 5.8 and 5.9(a) are also suitable for Table 5.10.

As the estimation results of coefficients and WTP values in the top half of Table 5.10 are

similar to the results in Table 5.8, the explanation would not be repeated here (see Table 5.8 for

the details). In this model, focus is placed on explaining class membership indicators. The lower

section in Table 5.10 presents the estimated parameters for the class membership indicators. The

parameters for the third class (―Verification Seekers‖) are all equal to zero due to their

normalization during estimation (Boxall and Adamowicz, 2002). Therefore, the other three

classes are described relative to the third class. Three exogenous covariates (income, education

and heart disease) and two key factors (attitude and trust) have been selected as the membership

indicators to explain the sources of heterogeneity across classes.

124

The coefficient for the membership indicator, Income, is negative and significant in class 1,

indicating that compared with the respondents in class 3 (Verification Seekers), lower income

respondents are more likely to fall into class 1 (Price Sensitive Consumers). In other words, the

respondents‘ average income is relatively lower in the first class of ―Price Sensitive Consumers‖

relative to the third class of ―Verification Seekers‖. The coefficients for Income are insignificant

in the other two classes (class 2 and class 4), which implies that income does not explain the

source of heterogeneity across the latent classes of 2, 3, and 4. Put differently, no obvious

income-different effects exist across the latent classes of ―Verification Seekers‖, ―Functional

Milk Believers‖, and ―Health Claims Rejecters‖. Unlike Income, Education have insignificant

parameter estimates in the latent class 1, 2 and 4, indicating that education levels do not explain

the source of heterogeneity across the four latent classes.

The membership indicator, Heart Disease, is negative and significant in class 1,

indicating that consumers with heart disease are more likely to belong to the third class of

―Verification Seekers‖ relative to the ―Price Sensitive Consumers‖. Thus, on average, consumers

with heart disease are more likely to be ―Verification Seekers‖ than ―Price Sensitive Consumers‖.

This membership indicator does not explain the source of heterogeneity across the other three

latent classes.

Attitude towards functional food represents factor 1 as described in table 5.1, with

negative and significant parameters in class 1 and 4. It captures the fact that respondents with

positive attitudes towards functional foods are more likely to belong to the third class of

―Verification Seekers‖, compared with the ―Price Sensitive Consumers‖ and the ―Health Claims

Rejecters‖. In contrast, this membership indicator does not explain the source of heterogeneity

between ―Verification Seekers‖ and ―Functional Milk Believers‖.

125

Trust (factor 2), is negative and significant in classes 1 and 2, indicating that respondents

who tend to have more trust in health claims, nutrition labels and new food products are more

likely to belong to the third class of ―Verification Seekers‖, compared with ―Price Sensitive

Consumers‖ and ―Functional Milk Believers‖. However, this indicator does not explain the

source of heterogeneity between ―Verification Seekers‖ and ―Health Claims Rejecters‖. The

constant parameter is positive and significant in the first and second latent classes, indicating that

there are some other unobservable variables not included in the model that could also explain the

source of heterogeneity across the latent classes.

In summary, seven discrete choice models have been examined: (1) a Conditional Logit

model and WTP for the base model; (2) a Conditional Logit model with interaction effects

between main variables; (3) a Conditional Logit Model with interaction effects for covariates and

key factors; (4) a Random Parameter Logit model and WTP for the base model; (5) a Latent

Class Model and WTP for the base model; (6) a Latent Class Model with interaction effects for

covariates and key factors; and finally (7) a Latent Class Model with class membership

indicators. These models have enable a systematic evaluation of the effects of different types of

labelling information on consumer attitudes, the influence of behavioural, attitudinal and

demographic factors, and the existence of heterogeneity within the sample population.

5.8 Conclusion

The growing market for functional foods, especially functional dairy products, provides a

potential opportunity to improve the health of Canadians and enable the development of a new

value-added food sector. With the growing interest among consumers in the link between diet

and health, and the credence nature of the health attributes in functional food products, labelling

plays a key role in consumers‘ choices.

126

This estimation results chapter has the objective of answering the research questions

related to labelling effects as described in Chapter 1. Seven different estimation models have

been used in this chapter to examine various relationships with respect to different types of

labelling effects on Canadian consumers‘ functional food choices. They are all discrete choice

models based on random utility theory, including the Conditional Logit (CL) model, the Random

Parameter Logit (RPL) model and the Latent Class Model (LCM). The CL model is basic but

essential to derive other advanced models in the family of discrete choice models.

The CL model has several major limitations as indicated in section 4.1. The RPL model and

the LCM have been adopted in this chapter to address the limitations of the CL model. They are

popular methods to explore the unobserved heterogeneity in choices. Significant improvements

in the goodness of fit were found in the RPL model and the LCMs, compared with the CL base

model. The estimation results of the RPL model in this study indicate that strong variations exist

in consumers‘ preferences for: whether the milk product is enriched with Omega-3; whether

there is a health claim or a heart symbol present on food labels; and whether those health claims

are verified by the government or a third party. The LCM assumes that respondents can be

intrinsically grouped into a number of latent classes. Four latent classes have been identified in

the LCMs presented in this chapter. What is more, the LCM with class membership indicators

offers an excellent way to incorporate and explain the sources of heterogeneity.

WTP estimates for the main attributes were also provided for each of the discrete choice

models. The WTP method provides a means of interpreting the estimated parameters and

identifying the monetary values associated with changes in those attributes. Given the

hypothetical nature of the choice experiment, the WTP estimates are most usefully interpreted as

relative measures of the strength of preferences. The WTP values conjointly considering main

127

effects and the interaction effects for key socio-demographic and attitudinal variables allow

further identification of the heterogeneity in consumers‘ preferences for the main labelling

attributes.

According to the results from the seven models, full labelling is superior to partial

labelling. Among the different types of full health claims, a risk reduction claim is the most

preferred health claim by respondents. The specific preference ordering of full labelling and

partial labelling is found to be as follows: on average, a risk reduction claim is preferred or at

least no less than a disease prevention claim; and a disease prevention claim is preferred to a

function claim or the use of a symbol (partial labelling). The results reveal that there is no

significant difference between a function claim, such as ―good for your heart‖ and partial

labelling in the form of a red heart symbol.

The rules surrounding the use of structure-function claims in Canada are somewhat more

onerous than in the United States. For example, the US does not require pre-approval of the use

of a structure-function claim. These results suggest that the use of a heart symbol on functional

food products in the Canadian market may be just as effective as a structure-function claim. Thus,

while consumers (and obviously functional food manufacturers) in Canada might benefit from

the extension of risk reduction claims or disease prevention claims to a wider range of food

products (albeit only where these claims are scientifically justifiable), it is not clear that there is

as much to be gained from a consumer‘s perspective by a relaxation of the rules around the use

of a structure-function claim since the same information can be conveyed through the use of a

symbol. Two caveats are important, however: first the analysis only examines heart health claims

and the use of a heart symbol – the connection between the two being fairly self-evident –

structure-function claims pertaining to other health conditions (such as digestive health) may be

128

more difficult to convey clearly through the use of visual imagery. Second, the issue of consumer

protection from fraudulent or misleading health claims has not been addressed in this study. The

use of structure-function health claims, or indeed visual imagery, to imply a health benefit that is

not present in a product obviously would not assist consumers in making informed decisions

about (genuinely) functional foods.

Verification of health claims appears to be important to those consumers for whom health

claims matter. The CL model shows that, on average, consumers responded positively to

verification of a health claim by either a government agency or a credible third party. The RPL

model shows that consumers‘ preferences for the assurance of government and third party

verifications of health claims are heterogeneous. While the results indicate that consumers‘

perceptions of government verification are relatively superior to third party verification, the

LCM results reveal a more nuanced picture, with evidence of heterogeneity in consumer attitudes

towards the source of verification, as has been found in other recent studies (see for example,

Innes and Hobbs, 2011 and Uzea, Hobbs and Zhang, 2010). In general though, verification by a

government agency (Health Canada) appears to be important to those respondents who also

tended to value the presence of Omega-3 in a milk product (classes 2, 3, and 4), while reactions

to third party verification (Heart and Stroke Foundation) are less consistent.

With respect to consumers‘ acceptance of the Omega-3 functional ingredient and the effects

of other attitudinal and behavioural factors, the results show that the presence of Omega-3 does

not have a universal appeal to all consumers; it appeals to segments of consumers. For example,

having Omega-3 in milk products produces more positive perceived utility for the consumers

who have higher levels of income. Omega-3 also receives more positive responses from female

consumers with more positive attitudes towards functional foods. These asymmetric responses

129

may be deeply rooted in consumers‘ personal, social and psychological differences. Behavioural

economic research, such as the reference-dependent effects derived from Prospect Theory in

psychology, might help to explain the reasons and source of these heterogeneous responses. The

following chapter focuses on examining the existence of reference-dependent effects in

consumers‘ functional food choices.

130

Chapter 6: Reference-Dependent Effects

in Consumers‟ Functional Food Choices

Psychology opens another door for this study to examine consumers‘ preference

heterogeneity in their choice decisions about functional foods. Originally developed from

psychology, Prospect Theory (Kahneman and Tversky, 1979) and the Reference-Dependent

Model (Tversky and Kahneman, 1991) have been widely applied in economic decision theory.

As discussed in Chapter 2, the information asymmetry inherent in functional foods creates

uncertainty for consumers about the presence/absence of a functional ingredient (e.g. Omega-3)

and about the health effects of functional foods. Labelling plays a key role in informing

consumers about health outcomes, therefore reducing uncertainty. However, different labelling

scenarios in the choice experiment reveal the actual/alternative product attributes (e.g. price or

Omega-3) which might be different from consumers‘ reference levels. The difference between

an actual and reference level of the attribute creates a so-called reference-dependent effect (Hu,

Adamowicz and Veeman, 2006). Reference-dependent effects are caused by consumers

weighing losses from a reference point more than equivalent sized gains (loss aversion)

(Tversky and Kahneman, 1991).

Following Chapters 4 and 5, this chapter continues to take Random Utility Theory as the

underlying theoretical framework and uses the stated preference survey data to examine the

existence of reference-dependent and loss aversion effects in Canadian consumers‘ functional

food choices. Building on the base model in Chapter 4, this chapter constructs utility functions

that incorporate reference-dependent effects and tests the existence of those effects. This helps

to extend and further our understanding of the overall effects of the price and Omega-3

attributes, two of the main attributes examined in Chapter 5. Furthermore, the extended models

131

in this chapter incorporate psychological characteristics to understand consumers‘

heterogeneous choices for Omega-3 functional milk. This chapter focuses on testing the

reference-dependent effects for a functional ingredient (e.g. Omega-3) and for price. The

Conditional Logit model is adopted to estimate these reference-dependent effects. The sources

of heterogeneity for the reference-dependent effects are also examined through the inclusion of

demographic variables and three attitudinal factors described in Chapter 4. This chapter includes

six sub-sections: (1) general introduction and background, (2) introduction of Prospect Theory,

(3) the Reference-Dependent Model and modelling reference-dependent effects, (4) data and

econometrics models, (5) estimation results and (6) conclusions.

6.1 Introduction

Consumer goods are generally categorized as having three types of attributes: search,

experience and credence (Nelson, 1970). The quality of a credence attribute cannot be

determined either before or after purchase (e.g. the nutritional component of the food).

Functional foods are usually considered to be foods with credence attributes. Since (in the

absence of labelling) the functional properties cannot be inferred by the consumer before or

even after purchase, consumers face uncertainty regarding the functional components of foods.

Even with the presence of labelling, information uncertainty is also apparent in the functional

foods market with respect to whether the functional health claim is credible, which depends on

whether consumers believe that consumption of functional foods leads to specific health

outcomes.

In neoclassical economic theory, expected utility theory is often used to analyze consumer

choice behaviour under uncertainty. In expected utility theory, human rationality is one of the

fundamental assumptions: it assumes that a consumer‘s rational choice should be consistent and

132

coherent. A great deal of evidence in the literature shows that individual consumers sometimes

behave ‗inconsistently‘ and ‗irrationally‘, which violates the predictions of expected utility theory.

In Prospect Theory, Kahneman and Tversky (1979) described decision problems that violate the

requirement of consistency and coherence, and they ―trace these violations to the psychological

principles that govern the perception of decision problems and the evaluation of options‖ (p.453).

Prospect Theory evaluates decisions between alternatives which involve risk, i.e. alternatives

with uncertain outcomes (e.g. gambling). In 1991, Tversky and Kahneman extended Prospect

Theory and proposed a consumer choice model incorporating loss aversion in riskless choice (e.g.

negotiations8), the so-called Reference-Dependent Model (RDM). The central assumption of the

RDM (also for Prospect Theory), is that losses and disadvantages have a greater impact on

preferences than gains and advantages (Tversky and Kahneman, 1991).

In the RDM, the reference-dependent effects are used to measure the value of changes

involving a gain or a loss compared with a reference point and the changes to the reference point

might lead to reversals of preference. The RDM and Prospect Theory evaluate the changes or

differences of attributes rather than the absolute magnitudes. For example, the attributes of

brightness, loudness or temperature, are evaluated by the past or present experience defined as a

reference point, and the values are perceived according to this reference point. Therefore, the

temperature could be evaluated as hot or cold depending on which reference level it has been

compared to. The same principle also applies to non-sensory attributes such as health, prestige

and wealth (Kahneman and Tversky, 1979).

8 Loss aversion can be involved in negotiations, since experimental evidence indicates that negotiators are less likely to achieve

agreement when the attributes over which they bargain are framed as losses than when they are framed as gains (Tversky and

Kahneman, 1991).

133

Think about the example of purchasing a milk product. First, consider yourself as a typical

conventional milk consumer. Suppose one morning you find that the conventional milk is used up.

When you go to the grocery store, you consider purchasing Omega-3 milk which is claimed to

have health benefits. In this case, will you evaluate the Omega-3 milk independent of your

typically consumed conventional milk? Will you consider the extra health ingredient (Omega-3)

as a gain in terms of improved product quality? On the other hand, to take the opposite case, if

you typically consume Omega-3 milk and consider buying conventional milk on this shopping

trip, will you evaluate this conventional milk independent of Omega-3 milk? Will you treat the

missing health ingredient (Omega-3) as a loss with respect to product quality? Would you have

the same intensity of feeling about the ‗gain‘ and ‗loss‘ of the Omega-3 in the above two cases?

Extensive empirical evidence suggests that losses will be weighed more heavily than the same

magnitude of gains, which is well-known as the property of loss aversion (Tversky and

Kahneman, 1991).

Some of the literature about modelling heterogeneous consumer choices focuses on taste

heterogeneity, and it might be tempting to make the behavioural assumption that all differences

in consumers‘ choices are reflected in their taste variations. This is likely to be an over

statement. At least, there might be some reference-dependent effects in the RDM that could be

used to explain part of the heterogeneity in choices (Hu, Adamowicz. and Veeman, 2006).

Reference-dependent effects reflect the fact that individuals with different reference points

could treat the same choice decision as a gain or a loss. A gain for one individual might be

treated as a loss by another. In the economics and marketing literature, reference-dependent

effects have been recognized for decades. A number of studies have focused on how to capture

and measure reference-dependent effects, such as Koszegi and Rabin (2006), Munro and

134

Sugden (2003), Suzuki, Tyworth and Novack (2001), Bell and Lattin (2000), Prelec (1998),

Ordonez (1998), Lattin and Bucklin (1989) and Winer (1986). Miljkovic (2005) conducted a

review of the hypothesis of perfect rationality regarding rational choice and irrational individual

behaviour. Putler (1992) developed a theoretical consumer choice model incorporating reference

price effects. Price reference effects have been examined in the marketing literature, e.g. Winer

(1988) and Briesch et al. (1997). Some researchers developed models of brand choice

incorporating price reference behaviours, e.g. Winer (1986), Kalwani et al. (1990) and Lattin

and Bucklin (1989).

Hardie, Johnson and Fader (1993) model loss aversion and reference-dependent effects in

brand choice. The authors argue that consumer choice is influenced by the position of brands

relative to multi-attribute reference points. They use the most recent brand purchased by each

household as the reference brand. They found that consumers perceive losses from a reference

point more than equivalent sized gains (loss aversion). Using scanner data, they generate a

Multinomial Logit Model to incorporate reference-dependent effects for price and quality

attributes. The developed model provides a better fit in both estimation and prediction than a

standard Multinomial Logit Model, and the model‘s coefficients demonstrate significant loss

aversion. Their model provides insights for the development of the econometric models in this

study.

Dhami and al-Nowaihi (2007) made a comparison between Prospect Theory and Expected

Utility Theory for the issue of people paying taxes. The authors argue that given relatively low

audit probabilities and penalties, why should people pay taxes. According to Expected Utility

Theory, evasion should be extremely attractive. So why do most taxpayers not evade? Expected

utility theory is unable to explain this. If people evade taxes, the loss will be the small

135

probability of being caught and charged a penalty. The results show that the magnitude of tax

evasion predicted by Prospect Theory is consistent with the data, since Prospect Theory

characterizes individuals as loss averse with respect to certain reference income and they

overweigh small probabilities while underweighting large ones. The authors conclude that

Prospect Theory can explain the tax evasion puzzle and the behaviour of taxpayers also provides

strong support for Prospect Theory.

Some researchers have examined reference-dependent effects for consumer promotions,

such as Kalwani and Yim (1992) and Lattin and Bucklin (1989). Kalwani and Yim (1992)

conducted an experimental study regarding price promotion and consumer price expectations.

The study examined the impact of price promotion on consumers‘ price expectations and brand

choice through an interactive computer-controlled experiment. The authors argue that price

promotion, by introducing a product at a lower than regular price, is shown to have an adverse

effect on subsequent sales, because consumers adopt the low promotion price as a reference and

consider the regular price as unacceptable and greater than the price they expect to pay. The

authors conclude that in the case of price expectations, promotion expectation losses loom larger

than gains, which is consistent with Prospect Theory. This study suggests that although offering

frequent price promotions can increase short term gains in sales, those sales increases may be

offset by the losses in sales caused by consumers not receiving the promotion prices they

expected.

The reference-dependent and loss aversion effects related to consumers‘ functional food

choices are examined in this chapter. In addition to the reference-dependent effect for price, this

chapter also examines the reference-dependent effect for the functional ingredient (e.g. Omega-

3), a credence characteristic. The sources of heterogeneity for these reference-dependent effects

136

are also examined. The following section provides an overview of Prospect Theory and the

Reference-Dependent Model.

6.2 Prospect Theory

As mentioned previously, in the standard consumer decision-making models, Expected

Utility Theory is typically used to model consumers‘ choice behaviors, in which the utility of a

choice is represented by the sum of the possible outcomes weighted by their probabilities.

Prospect Theory focuses on the changes in the current level of wealth and defines the changes as

gains or losses relative to a reference level. In contrast, Expected Utility Theory only focuses on

the final asset, and therefore only uses the ‗final value of change in wealth‘ as a main variable in

the utility function. In Prospect Theory (Kahneman and Tversky, 1979), the overall value

function of a prospect (V) is expressed as:

)()()()(),;,( yvqxvpqypxV (6.1)

Where v(.) is the value function which measures the value of deviations from the

reference point (i.e. gains or losses); v(x) and v(y) are associated with the subjective value of

outcomes x and y, respectively. It assumes that v(0) = 0, a basic property of the overall value

function. π is the weighting function which is associated with each probability of prospect; π(p)

and π(q) reflect the impact of the probability of p and q on the over-all value of the prospect,

respectively. However, the weighting function is not a conventional probability function, and it

measures the desirability of a prospect due to its possible outcomes. π is an increasing function

of probability p. π(0) = 0, and π(1) = 1 imply an impossible event and a must-occur event,

respectively. Another important property of the weighting function is that π(p)>p for low

probabilities and π(p)<p for high probabilities, but evidence shows that for all 0<p<1, π(p)+

137

π(1-p)<1 (Kahneman and Tversky, 1979). This property shows that individuals tend to

overreact to events with a smaller probability, and under react to events with larger probabilities.

See Kahneman and Tversky (1979), Laibson and Zeckhauser (1998), and Drazen (1998), for the

details of the summarized properties of the weighting function.

Notice that both the value function and the weighting function are subjective measures

and they vary across individuals. The value function is defined as gains or losses relative to a

certain reference point. The reference point is also a subjective measure, and each individual

might have a different level of reference which means they might have different views of gains

or losses relative to one reference point. Prospect Theory originally focused on the analysis of

decision making under uncertainty and was represented by an asymmetric S-shaped value

function as depicted in Figure 6.1, in which value is measured by gains or losses rather than

final assets. The value function includes two segments: the reference level which is treated as

the reference point, and the magnitude of the change (positive or negative) which depends on

the reference point. The following section discusses the S-shaped value function in the context

of the reference-dependent model.

6.3 The Reference-Dependent Model

Derived from Prospect Theory, the reference-dependent model develops a consumer

choice model involving loss aversion in riskless choice and measures the value of changes

involving a gain or a loss compared with a reference point; and changes in the reference point

might lead to reversals of preference. According to the psychological analysis of value,

reference levels play a large role in preference construction. People are more sensitive to an

outcome valued as a gain or a loss compared with a reference level, than the absolute final

138

outcomes. The central assumption of the model is that losses and disadvantages have a greater

impact on preferences than gains and advantages (Tversky and Kahneman, 1991).

Figure 6.1: An Illustration of a Value Function

Source: Tversky and Kahneman (1991)

In the reference-dependent model, outcomes are expressed as gains or losses from a

neutral reference outcome. As illustrated in Figure 6.1, the value function is commonly

asymmetrically S-shaped, concave above the reference point and convex below it. There are

three essential properties of the value function summarized by Tversky and Kahneman (1991,

p.1039): first, ―Reference dependence: the carriers of value are gains and losses defined relative

to a reference point.‖ Individuals‘ preferences depend on a reference point, and the changes of

reference point might lead to reversals of preference; second, ―Loss aversion: the function is

steeper in the negative than in the positive domain; losses loom larger than corresponding

gains.‖ Since losses dominate, people will work harder to avoid losses than to obtain gains; third,

―Diminishing sensitivity: the marginal value of both gains and losses decreases with their size‖.

139

For example, the difference in value between gains of $1 and $10 is greater than the difference

between gains of $101 and $110, similarly for the case of losses.

In the reference-dependent model, the value function Ri is expressed as follows:

iiiiii

iiiiii

ii

rxifruxu

rxifruxu

xR

)]()([*

)()(

)(

(6.2)

Where Ri(.) is the reference value function and ui(.) is the utility function which are

associated with the reference state ri. The first equation captures the reference gain effect and

the second equation captures the reference loss effect. is a parameter which can be described

as a coefficient of loss aversion with a restriction that is greater than 1. This expression

comes from the central assumption of the RDM, which is that losses and disadvantages have

greater impacts on preferences than gains and advantages.

Tversky and Kahneman (1991) concluded that if a choice could be viewed as a package of

product attributes, each attribute could be described by a value function specific to that attribute.

An individual could judge the value of each attribute based on his perceived reference point and

make a choice decision relevant to that reference point. In Random Utility Theory, a consumer‘s

utility of choosing a product is determined by the product‘s attributes and the utility is also a

function of the product‘s attributes. Thus, Random Utility Theory could be re-specified as the

value function by adding in reference-dependent variables in the reference-dependent model

(Tversky and Kahneman, 1991; Hu, Adamowicz and Veeman, 2006).

140

6.3.1 Modelling Reference-Dependent Effects

Standard consumer utility theory assumes that when making decisions, preferences do not

depend on current assets. Thus, in the case of a decision to consume a healthier food product,

standard consumer utility theory would not take into account a consumer‘s current situation.

This assumption greatly simplifies the analysis of consumers‘ choice decisions. According to a

psychological analysis of value, reference levels play a large role in determining preferences.

Understanding the determinants of consumers‘ perception of the health effects of functional

foods could provide insights into the extent to which consumers are likely to change their

consumption patterns in response to new information. The regulatory environment governing

allowable labelling claims affects consumers‘ reference points by altering the information set

available to consumers. Reference-dependent effects reflect the fact that individuals with

different reference points could make heterogeneous choice decisions. Thus, a utility function

need to be constructed that depends on the reference condition.

In modelling reference-dependent effects, a critical challenge is to identify the reference

point for each respondent. In the literature, particularly in experimental studies, reference points

are often presented or manipulated in the choice scenarios. For example, Kahneman, Knetsch

and Thaler (1990) used a decorated mug as a tool of manipulation in their experimental study.

They found that the monetary equivalent for the compensation for ―losing the mug‖ is greater

than the monetary value required to ―gain‖ the same mug. Hardie, Johnson and Fader (1993)

also argue that ―in experimental studies, reference points can be manipulated by changing the

status quo or by varying a target or norm‖ (P382). In their study, they used the most recent

brand purchased by each household as the reference point. They argue that the currently held

141

alternative could be the status quo and therefore a natural candidate for comparison. Suzuki,

Tyworth and Novack (2001) adopted a similar method in a transportation study.

The methods used in the above mentioned studies are very effective in clearly defining

the reference points when reference attributes are objective. Hu, Adamowicz and Veeman (2006)

further developed a method to define reference points when reference attributes are latent or

credent by considering respondents‘ perceptions for credence attributes of food products. In

their study, respondents‘ perceptions were determined by asking them whether the bread they

most often bought contained GM ingredients. This information was used as the reference point

for the GM attribute.

In this study, the reference-dependent effect of Omega-3 is a key variable to be

examined. Like GM ingredients in Hu, Adamowicz and Veeman‘s (2006) study, Omega-3 is

also a credence attribute for functional food products. Following the method used by Hu,

Adamowicz and Veeman (2006), this study establishes reference points by asking respondents‘

about the milk products they typically purchase.

According to Tversky and Kahneman (1991), the random utility function could be taken to

mimic the value function of the reference-dependent model. Based on ordinary Random Utility

Theory, E(u) captures the average expected utilities for all consumers. Consumers have different

expected utilities from a given product. For an individual consumer i, his utility function could

be expressed as:

),0()( 2 NuEu iii (6.3)

142

First, take the reference-dependent effect of price as an example, assuming that all

variables are constant in the utility function with the exception of price. The utility function for

consumer i choosing alternative j could be expressed as:

),0( 2

0 NPGPLPuu jjjjij (6.4)

where variable PLj captures the difference between the reference price and the observed

price when the observed price is above the reference price; PGj captures the difference between

the reference price and the observed price when the observed price is below the reference price;

variable Pj is the observed/actual price and the constant u0 captures a base level of utility from

consuming the product. The restriction that >1 captures the asymmetric response above and

below the reference point, which means the decision maker is loss averse. For empirical

applications, the test for the presence of loss aversion is that the estimated value of is

significantly greater than one (Bell and Lattin, 2000).

The gain and loss variables generated in this study follow the dummy variable method of

Hardie, Johnson and Fader (1993) and Hu, Adamowicz and Veeman (2006). This method

creates dummy variables that reflect ―gains‖ and ―losses‖ from the reference points.

Respondents in this study were asked: ―how much would you expect to pay for a 2-litre carton

of milk enriched with Omega-3?‖ That price information is taken as the reference point for price

which is the reference price (Pr) of Omega-3 milk. In the choice experiment, let the price of

Omega-3 milk in an alternative in each choice set be taken as the alternative/observed price (Pa).

Comparing Pr with Pa, if Pa < Pr, PGj = 1, otherwise 0; if Pa > Pr, PLj = 1, otherwise 0. If Pr – Pa

= 0, then PGj = PLj = 0. Based on these definitions of reference-dependent variables, price gain

= price loss = 0 (PGj = PLj = 0) in the fourth (no purchase) alternative in each choice set (Hu,

143

Adamowicz and Veeman, 2006). In this study, since the research question focuses on

consumers‘ reactions to the changes in the price of Omega-3 milk (not conventional milk), it

assumes that there is no price gain or price loss (PGj = PLj = 0) in the third alternative in each

choice set (the conventional milk alternative).

Respondents in this study were also asked: ―what is the price of the 2-litre carton of milk

you typically purchase?‖ This price information also has the potential to be used as the reference

point for price (both for Omega-3 and for conventional milk). Statistical analysis of the survey

data indicates that the variation is not statistically significant between this price information and

the price information used in the choice experiment. Therefore, the variation is quite small for

these two sets of price data. Testing also showed that no price gain or loss effect is observed if

this price information is taken as the reference point for price. Considering the pre-testing and

specific research questions of interest, respondents‘ expectation of the price they would expect

to pay for a 2-litre carton of Omega-3 milk is taken as the reference point for price in this study,

as explained above.

Second, consider a utility function with a reference-dependent effect for Omega-3

(functional ingredient). Assume all variables are constant in the utility function with the

exception of the Omega-3 and price attributes, thus the utility function could be expressed as:

jjjjjij LOmegaGOmegaOmegaPuu 3*330 (6.5)

Omega3Gj and Omega3Lj represent the reference-dependent gain and loss effects for the

Omega-3 attribute. In the survey, respondents were asked ―Is the milk you typically purchase

Omega-3 enriched milk?‖ This information provides the reference point for the Omega-3

attribute, Omega-3r. When a respondent perceives that the milk he or she typically purchases is

144

enriched with Omega-3, Omega-3r =1, otherwise Omega-3r = 0. In the choice experiment, the

actual presence of Omega-3 in a milk product is labelled (for the first two alternatives in each

choice set). When the milk product is explicitly labelled as containing Omega-3, Omega-3a = 1,

otherwise Omega-3a = 0. As mentioned previously, the presence of Omega-3 is a credence

characteristic both in terms of its presence (in the absence of labelling) and its efficacy. For the

convenience of terminology, it is defined that when Omega-3a – Omega-3r =1, Omega3Gj = 1

and Omega3Lj = 0; when Omega-3r – Omega-3a =1, Omega3Lj = 1 and Omega3Gj = 0. Again,

based on the definitions of reference-dependent variables and the particular choice experiment

design in this study, the fourth alternative (no purchase) in each choice occasion, Omega3Gj =

Omega3Lj = 0 for the Omega-3 attribute.

In this study, the usable sample size is 740 respondents. About 30% of respondents expect

to pay $2.69 or less for a 2-litre carton of Omega-3 milk. Respondents who expect to pay $3.59,

$4.49 or more for a 2–litre carton of Omega-3 milk account for 50%, 12% and 8% of the sample,

respectively. Regarding the presence of Omega-3 in their purchased milk products, a total of

5.3% of the respondents claimed that they typically purchase Omega-3 enriched milk. A total of

16.6% of the respondents claimed that they typically purchase milk enriched with a health

ingredient other than Omega-3, such as calcium or another ingredient. In this study, each

respondent was presented with eight choice sets, and each choice set contained three alternatives

described by different attribute combinations. In total, there are 17,760 ‗milk products‘ included

in this survey data. According to the definition of gains and losses for price and Omeag-3 as

defined above, 47% of the alternatives created a gain and 2.6% (about 500 alternatives) created

a loss for the Omega-3 attribute, while, 50.4% of the alternatives did not involve any gains or

losses for Omega-3. In terms of price, 14.6% and 34% of the alternatives involved gains and

145

losses for price, respectively, and the remaining alternatives (51.4%) did not involve any gains

or losses for price. These numbers suggest that there might be some gains or losses related to

reference-dependent effects for the price and Omega-3 attributes in this study.

Notice that one possible reference-dependent effect of a key attribute, health claim, is not

included in this study. As introduced in Chapter 1, unlike in the United States, explicit Omega-3

health claims (whether a function, risk reduction or disease presentation claim) are not currently

permitted on food labels or in advertisements in Canada. In the Canadian food market Omega-3

enriched milk cannot currently be labelled with a full health claim. In this study, it is assumed

that Canadian consumers have not consumed Omega-3 enriched milk with a health claim in

another country, such as the United States. Therefore, it is impossible to generate reference-

dependent variables for the presence/absence of a formal Omega-3 health claim, but would be

an interesting area for future research.

According to the notion of loss aversion in the reference-dependent model, people tend to

prefer to avoid losses over acquiring gains. Psychologically, losses are much more powerful

than gains. As explained, a reference-dependent effect for health claim could not be generated in

the same way as for the price and Omega-3 attributes. Recall the estimation results in Chapter 5

where it was concluded that consumers are more likely to prefer risk reduction claims and

disease prevention claims relative to function claims. One way to explain the results is by

consumer‘s taste variation as shown in Chapter 5, since risk reduction claims and disease

prevention claims are more concrete claims, while function claims might be more abstract. In

addition, the effect of message framing might provide another way to explain the results

(Tversky and Kahneman, 1981). In the health claim framing literature, Kleef, Trijp and Luning

(2005) mentioned that health claims may be formulated to focus attention on its potential to

146

provide a benefit or gain (e.g. function claim), or on its potential to prevent or avoid a loss (e.g.

risk reduction claim). Their results found that reduced disease risk-framed health claims have

significantly higher purchase intention ratings than enhanced function-framed health claims. To

some extent in this study, the function claim (―Good for your heart‖) might be taken as one way

to represent a gain effect from a health claim. The risk reduction claim (―Reduces the risks of

heart disease‖) and the disease prevention claim (―Helps to prevent coronary heart disease‖)

might be treated as a way to express loss avoidance in a health claim. The concept of loss

aversion from the reference-dependent model might provide another potential explanation of

this result. People are expected to be more sensitive to an outcome valued as loss avoidance than

as a gain. It is not the reality of loss that matters but peoples‘ perceptions. This suggests that an

individual‘s perception of a function claim phrased as a general health improvement (gain

effect), might be expected to have less impact on his/her preference than a risk reduction claim

or disease prevention claim which is phrased as loss avoidance. We expect that people will be

willing to pay higher to avoid losses than to obtain gains.

This section has modelled the reference-dependent effects for price and the presence of

Omega-3. The following section specifies the econometric models to measure the reference-

dependent effects for these attributes.

6.4 Data and Econometric Model

The data used in this chapter are obtained from the stated preference survey described in

Chapter 3. Due to the overlap between the data set used in this chapter and in Chapter 5, only the

unique part of the data which is related to the reference-dependent effects is mentioned here. To

measure the reference points for price and Omega-3, each respondent was asked how much he or

she expected to pay for a 2-litre carton of milk enriched with Omega-3, and whether the milk

147

product he or she typically purchases is Omega-3 enriched milk. Respondents‘ answers to those

questions are treated as their perceived reference levels for the price and Omega-3 attributes. In

the choice experiment, each product contains different levels of price and presence/absence of

Omega-3. Respondents‘ choices in each choice set were observed, and the choices that they

made were taken as the actual levels for the price and Omega-3 attributes. Since each respondent

completed 8 choice sets, each respondent has 8 observations. The difference between the

reference point and the actual observed choice is used to generate the reference-dependent

effects for the price and Omega-3 attributes for each respondent. Table 6.1 summarizes the

variable descriptions for the reference-dependent model.

Table 6.1: Variable Descriptions for the Reference-Dependent Model

Attributes Abbreviation Description

Price Price The price adopted in the choice experiment for a two-litre

carton of milk, ranging from $1.99 to $4.49.

Function Claim FC = 1 if the milk product has a function claim (‗Good for your

heart‘), otherwise 0.

Risk Reduction Claim RRC =1 if the milk product has a risk reduction claim (‗Reduces

the risks of heart disease and cancer‘), otherwise 0.

Disease Prevention

Claim

DPC =1 if the milk product has a disease prevention claim (‗Helps

to prevent coronary heart disease and cancer‘), otherwise 0.

Heart Symbol Heart = 1 if the milk product has a red heart symbol, otherwise 0.

Government

Verification

GOV = 1 if the health claim on the milk product is verified by

government (Health Canada), otherwise 0.

Third Party

Verification

TP = 1 if the health claim on the milk product is verified by a

third party (Heart and Stroke Foundation), otherwise 0.

Omega-3 Omega3 = 1 if the milk product contains Omega-3, otherwise 0.

No Purchase NoPurchase =1 if an alternative represents ‗not to purchase any milk

product‘ in the choice set, otherwise 0.

Reference-Dependent Variables

148

PG - = 1 if the alternative price involves a gain which means it is

below the respondent‘s reference price, otherwise 0.

PL - = 1 if the alternative price involves a loss which means it is

above the respondent‘s reference price, otherwise 0.

Omega3G - = 1 if the alternative involves a gain in Omega-3 attribute,

otherwise 0.

Omega3L - = 1 if the alternative involves a loss in Omega-3 attribute,

otherwise 0.

Exogenous Covariate Variables and Key Factors

Heart Disease HeartDisease =1 if a respondent self-reported that he or she has heart

disease, otherwise 0 (as defined in Table 5.1).

Income Income A demographic variable representing respondent‘s income,

as defined in Table 5.1.

Age Age A demographic variable representing respondent‘s age

Attitudes towards

functional food

Attitude Factor 1 from the factor analysis in table 4.2

Trust in health claims

and nutrition labels

Trust Factor 2 from the factor analysis in table 4.2

Health knowledge Knowledge Factor 3 from the factor analysis in table 4.2

With the specified attributes and levels of milk products in the choice experiment,

consumer i will choose his/her preferred alternative j in each choice set if and only if the utility

associated with alternative j is greater than other alternatives. According to Random Utility

Theory, the indirect utility of consumer i choosing alternative j can be expressed as the

following function:

Uij = ß1NoPurchase +ej (6.6)

Uij = (1-NoPurchase)*(ß2FunctionClaimj + ß3RiskReductionClaimj +

ß4DiseasePreventionClaimj + ß5HeartSymbolj + ß6Govj + ß7ThirdPartyj

149

+ß8Omega-3j + ß9Pricej + ß10PriceGainj + ß11PriceLossj + ß12Omega3Gainj +

ß13Omega3Lossj) + ej (j ≠ no purchase) (6.7)

Most variables in equation (6.7) are described in Chapter 4 and will just be briefly repeated

here. NoPurchase is a dummy variable that equals 1 if an alternative represents ‗not purchase any

of the milk products‘ in the choice set and equals 0, otherwise. Omega-3 is an alternative specific

constant variable equal to 1 if the milk product includes Omega-3 ingredient and equal to 0,

otherwise. The health claims attribute is separated as four dummy variables: Function Claim,

Risk Reduction Claim, Disease Prevention Claim and No Claims. The verification organization

attribute is represented by three dummy variables, Government, Third Party and None. Heart is a

dummy variable equal to 1 if the milk product has a red heart symbol, otherwise 0. Price is the

retail price of the milk product in alternative j. ßs are the estimated parameters and ej is the error

term.

As defined in section 6.3.1, Omega3Gj and Omega3Lj are two dummy variables which

represent whether the choice involves a gain or a loss with respect to the presence of the Omega-

3 attribute in a choice option. In equation (6.7), Omega3Gainj is an interaction variable between

Omega3Gj and Omega-3j, because the gain effect for Omega-3 can only occur when the actual

option contains the Omega-3 ingredient. Similarly, Omega3Lossj captures the interaction

between the Omega3Lj and the third alternative specific constant which represents the absence of

the Omega-3 attribute, because the loss effect for Omega-3 can only arise when the actual option

does not contain Omega-3. For the price attribute, PGj and PLj are the gain and loss effects for

the price of Omega-3 milk which are also described in section 6.3.1. PriceGainj and PriceLossj

are two interaction variables that interact variables PGj and PLj with the actual price to capture

the reference-dependent effects for the price attribute. According to the loss aversion prediction

150

of the reference-dependent model, the value function should be steeper in the loss domain than

the gain domain since losses are expected to have a larger impact on consumers‘ preferences

than gains. To be consistent with the prediction, the magnitude of the coefficient for PriceLossj is

expected to be greater than the coefficient for PriceGainj. Similarly, the magnitude of the

coefficient for Omega3Lossj is expected to be greater than the coefficient for Omega3Gainj.

According to the prediction, the coefficients for Omega3Gainj and PriceGainj should be positive

while the coefficients for Omega3Lossj and PriceLossj should be negative.

Equation (6.8) specifies a random utility function including interaction effects between

the reference-dependent variables and some exogenous demographic and attitudinal variables.

These interaction effects help to test consumers‘ heterogeneous responses to the reference-

dependent variables. The indirect utility function with interaction effects of consumer i choosing

alternative j can be expressed as the following function:

Uij = (1-NoPurchase)*(ß2FunctionClaimj + ß3RiskReductionClaimj +

ß4DiseasePreventionClaimj + ß5HeartSymbolj + ß6Govj + ß7ThirdPartyj

+ß8Omega-3j + ß9Pricej + ß10PriceGainj + ß11PriceLossj + ß12Omega3Gainj +

ß13Omega3Lossj +γnZn*Xj) + ej (6.8)

In this equation, the main effect variables and the reference-dependent variables for the

price and Omega-3 attributes are the same as in equation (6.7). Where Zn represents the selected

exogenous covariate variables (e.g. income, education, heart disease and three key factors from

the Factor Analysis); Xj represents the reference-dependent variables for Price and Omega-3 that

can be interacted with Zn; γn represents the coefficients of those interaction variables. For

151

example, Zn*Xj can be specified as interaction variables, such as Incomei*PriceLossj,

HeartDiseasei*Omega3Gainj, Attitudei*Omega3Gainj, etc.

The five covariates (Income, HeartDisease and three key factors) were considered to

have the potential to explain the source of the reference-dependent effects examined in this

study. The following prior beliefs are informing the choice of variables: higher income

consumers are expected to be less price sensitive than those consumers with lower incomes.

Respondents who suffer from heart disease are expected to be more concerned with obtaining an

Omega-3 gain than those who do not have heart disease. Consumers who have positive attitudes

towards functional foods, with more health knowledge or who tend to trust health claims or

nutrition labels, are assumed to be more sensitive to obtaining gains or ‗suffer‘ more from losses

with respect to the presence of Omega-3 than those who do not have positive attitudes towards

functional foods, have less health knowledge or tend to be trusting of health claims or nutrition

labels.

This section has specified two models of the random utility function with reference-

dependent effects. The next section presents the estimation results for those models. The

Conditional Logit model is used to obtain estimation results.

6.5 Estimation Results

Table 6.2(a) presents the estimation results for two Conditional Logit models with

reference-dependent effects. The first CL model presents the results according to equation (6.7),

which contains both reference-dependent variables and main variables. The second CL model

presents the results according to equation (6.8), which includes the reference-dependent variables,

the main variables and the interaction variables. Table 6.2(b) presents the WTP estimates for the

152

results shown in Table 6.2(a). It should be noted that an RPL model including reference-

dependent effects was also estimated. Since the RPL model‘s results are similar to the CL model

in Table 6.2(a), the RPL estimates are presented in the Appendix 2, and the discussion in this

chapter focuses on the CL model results.

Comparing the estimation results for the first CL model in Table 6.2(a) with the base model

in Table 5.2 in Chapter 5, the only difference between those two models is the inclusion of

reference-dependent effects in Table 6.2(a). The values of the log likelihood functions are -

5146.678 in Table 6.2(a) and -5275.028 in Table 5.2, and the Pseudo-R2 are 0.268 and 0.25,

respectively. According to the LR test, there is a significant improvement in model fit for the CL

model in Table 6.2(a).

The estimated main effects in the CL model in Table 6.2(a) and Table 6.2(b) are similar to

the results in Table 5.2 in Chapter 5. As mentioned in Chapter 5, all the coefficients for the main

variables are significant at 1% and with the expected signs. On average, respondents prefer all

three types of full health claims relative to no health claim (base level). They also prefer a heart

symbol to be present on food labels relative to no symbol. They positively respond to the

verifications of health claims by a government agency or a third party. They also prefer milk

products enriched with Omega-3, but they dislike not purchasing any milk product or paying a

higher price. The discussion of results in this chapter will focus on the reference-dependent

effects.

153

Table 6.2 (a): Conditional Logit Estimates with Reference-Dependent Effects

CL Model CL Model

with Interaction Effects

Variable Coefficient t-ratio Coefficient t-ratio

PriceGain -0.044 -1.022 -0.044 -0.979

PriceLoss -0.187*** -4.899 -0.256*** -5.905

Omega-3Gain -0.542* -1.697 -0.501 -1.565

Omega-3Loss -1.673*** -4.555 -2.117*** -4.644

Price -1.243*** -29.076 -1.327*** -29.874

Function Claim 0.323*** 3.095 0.313*** 2.935

Risk Reduction Claim 0.721*** 6.926 0.738*** 6.927

Disease Prevention Claim 0.568*** 5.553 0.588*** 5.613

Heart Symbol 0.196*** 4.221 0.205*** 4.328

Government Verification 0.377*** 5.710 0.418*** 6.150

Third Party 0.310*** 4.740 0.344*** 5.100

Omega-3 0.975*** 2.802 0.951*** 2.716

Income*PriceLoss - - 0.001*** 4.670

HeartDisease*Omega-3Gain - - 0.593*** 4.325

Attitude*Omega-3Gain - - 0.661*** 19.254

Trust*Omega-3Gain - - 0.271*** 8.366

Knowledge*Omega-3Gain - - -0.108*** -3.353

Attitude*Omega-3Loss - - -0.427 -1.513

Trust*Omega-3Loss - - 0.130 0.480

Knowledge*Omega-3Loss - - 0.902*** 2.851

No Purchase -5.625*** -43.239 -5.864*** -43.458

Log Likelihood Function of CL Model = -5146.678; Pseudo-R2 = 0.268

Log Likelihood Function of CL Model with Interaction Effects = -4873.861; Pseudo-R2 = 0.307

*,** and *** indicate significant at the 10%, 5% and 1% level, respectively.

154

Table 6.2 (b): WTP Estimates with Reference-Dependent Effects

CL Model CL Model

with Interaction Effects

Variable WTP

($/2 litres)

t-ratio WTP

($/2 litres)

t-ratio

PriceGain - - - -

PriceLoss - - - -

Omega-3Gain -0.436* -1.695 ^ ^

Omega-3Loss -1.346*** -4.511 ^ ^

Price - - - -

Function Claim 0.260*** 3.077 0.236*** 2.920

Risk Reduction Claim 0.579*** 6.819 0.556*** 6.833

Disease Prevention Claim 0.457*** 5.478 0.443*** 5.544

Heart Symbol 0.158*** 4.167 0.154*** 4.276

Government Verification 0.303*** 5.578 0.315*** 6.005

Third Party 0.250*** 4.708 0.259*** 5.070

Omega-3 0.784*** 2.760 0.717*** 2.676

^Omega-3Gain :

1) With heart disease, average attitude,

knowledge and trust levels.

2) Without heart disease, and with

average attitude, knowledge and

trust levels.

-

-

0.069

-0.378

0.267

-1.563

^Omega-3Loss:

With average attitude, knowledge,

and trust levels.

-

-

-1.596***

-4.602

No Purchase -4.524*** -48.379 -4.420*** -51.321

*,** and *** indicate significant at the 10%, 5% and 1% level, respectively.

“^” indicates that the WTP estimates have been moved to the last part of the table

The reference-dependent effects for Price and the presence of Omega-3 are measured in the

first CL model in Table 6.2(a). The coefficient for Price is negative and significant, which is

consistent with expectations. The coefficient for PriceGain is not significant, indicating that

consumers do not value the gain on price very much, perhaps because milk products are

relatively cheap and affordable for most consumers. However, the coefficient for PriceLoss is

155

negative and highly significant, indicating that the effect of a price loss has a significant and

negative impact on consumers‘ utility. Both the effects of price gain and price loss should be

included in the overall effect of price. The magnitude of the coefficient for PriceLoss is greater

than for PriceGain. The findings of the reference-dependent effects for price are consistent with

the properties of Prospect Theory: the consumer‘s value function has a different slope in the gain

than in the loss domain; they react differently in those domains, which means consumers are

concerned with losses more than gains.

The reference-dependent effects for Omega-3 are also measured in the first CL model in

Table 6.2(a). The coefficient for Omega-3 is positive and significant, which implies that the

presence of Omega-3 could increase the probability of consumers purchasing the product. The

coefficient for Omega-3Gain in this model appears to be negative and borderline significant at

the 10% level. This is unexpected and one potential explanation might be that Omega-3 gain

effect captures consumers‘ responses to the presence of Omega-3 when it is not expected. In this

study, 47% of the respondents involve Omega-3 gains in the purchase simulation, which means

those consumers claimed that they typically consume conventional milk but switch to Omega-3

milk in the choice experiment. Their responses to the unexpected presence of Omega-3 might be

heterogeneous. The negative coefficient of Omega-3Gain indicates that on average, the

unexpected presence of Omega-3 has a negative impact on consumers‘ utility. However, notice

that the combined effects of variable Omega-3 and Omega-3Gain are still positive, indicating

that the overall effect of the Omega-3 attribute still has a positive impact on consumers‘ utility.

The coefficient for Omega-3Loss is negative and highly significant at 1%, indicating that

respondents tend to discount Omega-3 losses in their utilities, and prefer not losing this attribute

156

if the milk products they typically consume already contained Omega-3. This result is generally

consistent with Prospect Theory: losses have larger impacts on consumers‘ utilities than gains.

Figure 6.2 presents a graphical interpretation of the reference-dependent effects

according to the results from the first CL model in Table 6.2(a) and the base model in Table 5.2.

In graph (a) and graph (b) in Figure 6.2, the dashed lines represent the coefficients of attributes

(Price and Omega-3) when reference-dependent effects are ignored, as per Table 5.2. To make

an easier comparison, note that the coefficient of Price is taken as its absolute value. The solid

lines represent the results when reference-dependent effects are considered as in Table 6.2(a).

The dotted lines represent the S-shaped value function illustrated by Prospect Theory.

157

Figure 6.2: Reference-Dependent Effects for Price and Omega-3

158

Table 6.2(a) also presents a CL model with interaction effects between the reference-

dependent variables and exogenous variables. The value of the log likelihood function is -

4873.861 and the Pseudo-R2 is 0.307, which according to the LR test, suggests a significant

improvement in the goodness of fit over the first CL model in Table 6.2(a). All the coefficients

of the main effects in the second CL model have similar values and significant levels to the

previously reported CL model in Table 6.2(a). Therefore, the interpretations of the main effects

are similar to the first CL model. The reference-dependent effects in the second CL model need

to be considered by the conjoint effects of the reference-dependent variables and their interaction

terms. The specific interpretation is follows.

The interaction effects in the second CL model of Table 6.2(a) need to be interpreted in

combination with the relevant reference-dependent variable. As indicated above, the coefficient

for PriceLoss is negative and highly significant, indicating that the effect of a price loss has a

significant negative impact on consumers‘ utility. The coefficient for Income*PriceLoss is

positive and significant, which reduces the overall magnitude of the price loss effect (PriceLoss

and Income*PriceLoss). Respondents with higher incomes are less likely to be hurt by a price

that is higher than their reference price, which means they are less price sensitive than lower

income consumers.

Recall that the coefficient of Omega-3Gain is insignificant in the CL model with

interaction effects, indicating that consumers do not value this attribute in general. However, the

coefficient for HeartDisease*Omega-3Gain is positive and significant, which increases the

overall magnitude of the Omega-3 gain effect (Omega3Gain and HeartDisease*Omega-3Gain).

Respondents with heart disease are more likely to respond positively to an Omega-3 gain, and

prefer the presence of Omega-3 for a two-litre carton of milk, when they currently do not

159

consume Omega-3 milk. Therefore, respondents with heart disease might believe that Omega-3

could generate a health gain, but they don‘t currently consume Omega-3 milk.

The coefficients for Attitude*Omega3Gain and Trust*Omega3Gain are all positive and

significant, which all increase the overall magnitude of the Omega-3 gain effect. While, the

coefficients for Knowledge*Omega3Gain is negative and significant, which decrease the overall

magnitude of the Omega-3 gain effect. As discussed in Chapter 4, ‗Attitude‘, ‗Trust‘ and

‗Knowledge‘ are the three key attitudinal factors from the Factor Analysis. Respondents who

have positive attitudes towards functional foods, more trust in health claims and nutrition labels

or with less health knowledge, tend to positively respond to the Omega-3 gain effect. Or in other

words, although these respondents currently might not consume Omega-3 milk, they are likely to

believe the health benefits of Omega-3. These respondents are more sensitive to the Omega-3

gain effect relative to those who do not have positive attitudes towards functional foods, who

tend to be less trust of health claims and nutrition labels, or who have more health knowledge.

Note that as discussed in Chapter 5, respondents with more health knowledge in this study tend

to be more scepticism toward Omega-3 milk, and this finding is consistent with some health

communication literature, such as Jallinoja and Aro (2000).

However, according to the WTP estimates in Table 6.2(b), the WTP values for Omega-3

gain are not significant, indicating that regardless of whether respondents have heart disease,

those with average attitudes towards functional foods, average health knowledge and trust levels

towards health claims and nutrition labels, are not willing to pay a significant amount of money

for the Omega-3 gain attribute. Therefore, although the coefficient of

HeartDisease*Omega3Gain is positive and significant, WTP estimates indicate that these effects

might not be significant enough to translate into dollar values.

160

As mentioned above, the coefficient for Omega-3Loss is negative and significant, which

implies that the Omega-3 loss effect has a negative impact on consumers‘ utility. The coefficient

for Knowledge*Omega-3Loss is positive and significant, which decreases the overall magnitude

of the Omega-3 loss effect and makes it less noticeable. Respondents with more health

knowledge seem less likely to believe the health benefits of Omega-3. Therefore, they are less

sensitive to, and tend to ‗suffer‘ less from, the absence of Omega-3 when they usually purchase

Omega-3 milk. The coefficients for Attitude*Omega-3Loss and Trust*Omega-3Loss are

insignificant, which do not affect the overall magnitude of the Omega-3 loss effect. These results

indicate that respondent‘s attitudes towards functional food or trust in health claims or nutrition

labels seem not able to explain consumer‘s heterogeneous response to Omega-3 loss effect.

Nevertheless, the WTP estimates in Table 6.2(b) show that respondents who have

average attitudes towards functional foods, average health knowledge and trust levels towards

health claims and nutrition labels, would need to be compensated $1.60 to lose the Omega-3

attribute if they regularly consume milk products enriched with Omega-3.

6.6 Conclusions

Prospect Theory and the reference-dependent model (RDM) modify Expected Utility

Theory under uncertainty to accommodate apparently ‗irrational‘ but ‗real‘ observations about

human behaviour. In the RDM, the reference-dependent effect is used to measure the value of

change involving a gain or a loss compared with a reference point. This chapter examines the

reference-dependent effects for ‗Omega-3 gain or loss‘ and ‗price gain or loss‘ based on Prospect

Theory and the reference-dependent model. This chapter constructs utility functions including

reference-dependent effects and tests the existence of those effects. The analysis finds evidence

for the existence of reference-dependent and loss aversion properties. It suggests that the

161

reference-dependent effects for price and Omega-3 exist and are empirically verifiable.

Consumers‘ choice decisions could be influenced by the differences between reference points

and alternative observations. The results show that the reference-dependent effects for price gain,

price loss, and Omega-3 loss are all directly consistent with Prospect Theory. However, the

results for the Omega-3 gain effect are somewhat more complicated, requiring the inclusion of

interaction effects in the estimation model to be consistent with Prospect Theory. Reference-

dependent effects and loss aversion are evident for the price and Omega-3 attributes in this study.

As predicted by Prospect Theory, the losses tend to have a greater impact on consumers‘

preferences than the gains.

The results presented in this chapter differ from previous studies in that the source of

reference point effects are explained through consumers‘ demographic and attitudinal

characteristics by adopting interaction effects in the Conditional Logit model. Income,

HeartDisease, Attitude, Knowledge and Trust (three key factors) are used to explain consumers‘

heterogeneous responses to the reference-dependent effects for price and Omega-3. Income and

whether an individual suffers from heart disease are found to be two important personal

characteristics to explain consumers‘ diversified choices. In general, and unsurprisingly,

consumers with higher incomes are found less likely to suffer from a price loss. Consumers who

suffer from heart disease are more likely to obtain an Omega-3 gain effect from the milk

products, which indicate they pay more attention to health impacts relative to those who do not

suffer from heart disease. Attitudes towards functional foods, health knowledge and trust in

labels (three key factors) can help to explain the sources of heterogeneity surrounding the

reference-dependent effects.

162

The loss aversion property of the RDM opens another door to study the functional foods

market. According to the psychological principle of loss aversion, promotional messages for

functional foods could emphasize disease risk reduction rather than general health benefits,

since losses tend to dominate and people are expected to work harder to avoid losses than to

obtain gains. We can therefore expect the food industry to favour the use of stronger health

claims. This has two key implications for the regulatory environment governing allowable

health claims: the extent to which consumers are expected to adopt functional foods in response

to health claims, and second, the incentives that firms have to adopt health claims which imply

disease risk reduction functions. Government regulatory agencies should be aware that health

claims containing health information related to gains or loss avoidance might have distinct

influences on consumers‘ motivations to purchase functional foods. Therefore, agencies

regulating health claims need to be careful to control what types of health claims are permissible

on food labels. At the same time, the regulatory agency should also be aware that because of

consumers‘ asymmetric responses to health information including gains and loss avoidance,

food companies have strong incentives to lobby the government to allow health claims to imply

disease risk reduction functions. Therefore, tight regulation of allowable claims is very

important in protecting consumers.

A limitation of this study is that a relatively small proportion of respondents incur an

Omega-3 loss in this study. However, this limitation might also be explained as an important

finding or implication to the food industry, since a relatively small portion of consumers who

currently consume Omega-3 milk are very hard to convert back to consume conventional milk

anymore. Consumers seem to be loyalty to consume Omega-3 milk if they take Omega-3 milk

as their reference points. Further research might help to test these results, particularly if there are

163

more consumers taken Omega-3 milk as their reference levels in future. It will be interesting to

see in the long run, the implications for Canadian consumers‘ functional food choices when

their reference points for functional milk change.

164

Chapter 7: Implications and Conclusions

It has been argued that consumers‘ growing interest in functional foods provides value-

added growth opportunities to the Canadian agri-food sector (Agriculture and Agri-Food Canada,

2009). Consumers‘ response to functional foods is a relatively new research area with many

unanswered questions regarding public awareness of the health-enhanced properties of functional

foods. Given the credence nature of functional foods, labelling plays a key role in helping

consumers making informed consumption choices. This study has examined the effects of

labelling and verification of health claims through an analysis of consumers‘ stated preferences

for a specific functional food product, Omega-3 milk. In addition to consumers‘ perceptions of

the credibility of health claims verified by different organizations, some other potential

influences on consumers‘ decisions were also considered, such as attitudes towards functional

foods, consumers‘ health status and knowledge, trust in labels per se, socio-demographics and

the existence of reference-dependent effects for key attributes.

This research is one of the first studies to examine the effects of different types of health

claim labelling simultaneously. Specifically, this study investigated function claims, risk

reduction claims and disease prevention claims, in combination with other ways of signalling

health benefits, such as the use of a heart symbol, on Canadian consumers‘ functional food

choices. This study contributes to the knowledge in this domain by providing a comparative

analysis of different types of labelling strategies. The extant knowledge of labelling effects in the

formats of risk reduction claims, disease prevention claims and symbols on functional foods is

limited. One of the primary contributions of this study is addressing this gap in the literature.

There is also limited research regarding the credibility verification of health claims for functional

food, especially at the national level within Canada. To enrich the knowledge in this area, this

165

study examines the verification effects of different organizations for the credibility of health

claims across Canada, including a government organization (e.g. Health Canada) and a third

party (e.g. the Heart and Stroke Foundation). Another contribution of this study is to extensively

apply the psychographic segmentation analysis into the functional food sector, by using both

attitudinal and socio-demographic factors to identify and explain different segmentations of

Canadian functional food consumers. The findings of this study might be particularly useful to

the functional food industry and Canadian regulators, to help understand consumers‘ perception

of functional food labels, developing marketing strategies for different consumer segments, and

establishing an effective regulation system of health claims for the Canadian functional food

sector.

This chapter includes three sections: (1) summary of major research findings, with respect

to full labelling and partial labelling, effects of different verification organizations, the role of

attitudinal and demographic information, the reference-dependent effects and choosing a proper

discrete choice model; (2) implications for the functional food industry and public policy; (3)

limitations of this thesis and avenues for future research.

7.1 Summary of Major Research Findings

7.1.1 Full Labelling and Partial Labelling

For the government regulatory agency that decides which types of health claims for

functional foods are permissible, it is important to sufficiently understand how each type of

health claim affects consumers‘ choice decisions. In Canada, the regulatory environment

governing health claims for functional foods is somewhat more restrictive relative to some other

countries, such as the United States. While arguably this may offer more protection to consumers

166

from misleading health claims, it also means that food companies, as a consequence, often resort

to visual imagery, such as a red heart symbol, to imply certain health benefits. This type of

labelling strategy has been referred to ‗partial labelling‘ in this study. Correspondingly, ‗full

labelling‘ refers to explicating formal health claims on food labels, ranging from function claims,

to risk reduction claims, to disease prevention claims. Note that ―full labelling‖ refers to explicit

health claims on food labels and ―partial labelling‖ means to use an image or visual cue to

implicitly signal a health benefit. In this study, data was collected from an online survey

including a choice experiment administrated to 740 usable respondents across Canada in the

summer of 2009. This thesis uses discrete choice modelling to examine the participants‘

responses to different labelling strategies (full and partial) for milk enriched with Omega-3.

The results indicate that full labelling is strongly preferred over partial labelling,

especially for risk reduction claims. The preference ordering for health claims (full labelling)

reveals that, on average, a risk reduction claim is preferred or at least no less than a disease

prevention claim by most respondents and a disease prevention claim is preferred relative to a

function claim. While certain types of risk reduction claims are permissible in food products in

Canada and the United States, disease prevention claims are not. There is no significant

difference in consumers‘ preferences between a function claim (such as ―Good for your heart‖)

and partial labelling in the form of a red heart symbol. Although a risk reduction claim for

Omega-3 is already allowed in the United States, such a claim has not yet been approved in

Canada. This study finds that Canadian consumers‘ responses to a risk reduction claim on

Omega-3 milk are quite positive, especially when verified by a government agency. Therefore,

an approval of risk reduction claims for Omega-3 functional ingredients might deserve a careful

consideration by the Canadian regulatory agency, assuming that there is sufficient scientific

167

evidence to support such a claim since it remains important to balance the objectives of

informing consumers with protecting them from misleading information.

7.1.2 Effects of Different Verification Organizations

Establishing the credibility of health claims is another challenge faced by the functional

food industry. Health claims need to be credible to the consumers to be effective as a marketing

tool. The credence characteristic inherent in a functional attribute makes it difficult for

consumers to evaluate the accuracy of health claims. Verification of claims by independent

organizations provides an opportunity for the functional food manufacturers to bolster the

credibility of the health claims on their products. This thesis examines the credibility of health

claims verified by different organizations, including a government organization (e.g. Health

Canada) and a third party certification (e.g. the Heart and Stroke Foundation). The results show

that government or third party verifications are preferred by the respondents relative to no

verification. The respondents showed more trust in the verification made by a government

agency, or at least no less, than the third party in different WTP estimates for verifications in

different discrete choice modelling analysis (the RPL model and the CL model). The Latent

Class Model results reveal a more nuanced picture, with evidence of heterogeneity in consumer

attitudes towards the source of verification. More specifically, verification by a government

organization appears to be important to those respondents who also tended to value more highly

the presence of Omega-3 in a milk product, while reactions to third party verification are less

consistent.

168

7.1.3 The Role of Attitudinal and Demographic Information

Another major finding of this study is that consumers‘ prior attitudes and their socio-

demographic status helps explain their choice decisions and should be included in research

assessing consumer preferences. For example, the results show that consumers with positive

attitudes towards functional foods are more likely to respond positively to the presence of

Omega-3 or risk reduction claims relative to those who have negative attitudes towards

functional foods. Consumers‘ health knowledge and their trust in health claims and nutrition

labels are also found to influence their choices for functional foods. Furthermore, consumers‘

health status and key socio-demographic attributes, such as income and education, also help to

explain their choice decisions. For food companies wishing to better understand how to

communicate health benefits to target consumer segments, these dimensions of the consumer

attitudes are important considerations.

This study finds that respondents with higher education levels tend to respond negatively

to the presence of a risk reduction claim. One possible explanation is that better educated

consumers might have greater levels of scepticism towards risk reduction claims. Actually, this

finding is consistent with marketing literature in the area of scepticism and advertising, such as

DeLorme, Huh and Reid (2009). The implication of this finding is that, functional food

companies may need to provide more evidence to reduce the scepticism of higher educated

consumers, such as verification by a credible third party or public agency.

The results of the LCM with class membership indicators further identify the source of

heterogeneity among different consumer segments. For example, compared with the Price

Sensitive Consumers (class 1), consumers who are Verification Seekers (class 3) tend to have

169

higher incomes, have suffered from heart disease, have generally more positive attitudes towards

functional foods and are more likely to trust health claims and nutrition labels.

The descriptive analysis of the survey data shows that: (1) ‗food labels‘ receives the

highest score for being a frequently used health information source, and respondents also claim a

higher trust level in food labels (one of the top 3 most trusted sources), indicating the importance

of food labels to consumers‘ healthy food choices; (2) although ‗internet‘ and ‗media‘ receive

higher response rates for being frequently used health information sources (among the top 3 most

frequently used sources), consumers‘ trust levels in those two sources are relatively lower.

Consumers might perceive that the internet and the media are less reliable information sources

but are easily accessible; (3) food companies/industry receives the lowest scores both for being a

frequently used health information source and for the trust in that information source. This

suggests that the functional food industry needs to enhance its communications with consumers

via labels and improve the credibility of health claims through government or third-party

verifications. The influence of health professionals is another way helping the food industry to

communicate with consumers.

7.1.4 Reference-Dependent Effects

In the reference-dependent model, consumers‘ utilities are assumed not only to depend on

the absolute value of attributes, but also on changes around these attributes‘ reference points, so-

called reference-dependent effects. The empirical model developed in this study incorporates

reference-dependent effects into classical random utility theory to explain consumers‘ judgment

and decision making behaviour in the context of functional food choices. The results indicate

that consumers‘ utilities could be changed if the price for Omega-3 milk diverges from

consumers‘ reference levels. The directions of these changes are different depending on whether

170

it occurs as a gain or a loss. In this study, as expected, price loss is observed to have a negative

impact on consumers‘ utility, while price gain does not have an obvious impact. The magnitude

of the price loss coefficient is larger than the price gain, consistent with the properties of the

reference-dependent model.

In addition to the ‗price gain and loss effects‘, this study also examined a second type of

reference-dependent effect - the ‗Omega-3 gain and loss effects‘. Omega-3 loss is found to have

a negative impact on consumers‘ utility. The impact of Omega-3 gain, however, is somewhat

more complex in this study. For example, in the base CL model, consumers‘ responses are

significant and negative towards the Omega-3 gain effect, as is also the case for the RPL model

in Appendix 2. However, when considering interaction effects with socio-demographic and

attitudinal variables in another CL model, the total effect of Omega-3 gain is no longer

significant. Therefore, the estimates of the reference-dependent effect for Omega-3 are less

consistent with the prediction of Prospect Theory.

The results of the conjoint effects indicate that overall consumers do not value the

presence of Omega-3 in the milk products if their typical consumed milk does not contain

Omega-3, however, for those consumers who have higher incomes, suffer from heart disease,

have positive attitudes towards functional foods, have lower levels of health knowledge or are

more likely to trust health claims and nutrition labels, they tend to respond positively to the

Omega-3 gain effect. Generally speaking, the inclusion of reference-dependent effects in choice

models offers a better explanation of consumers‘ choice behaviours by considering the effect of

key attributes diverging from a consumer‘s reference point.

171

7.1.5 Choosing a Proper Discrete Choice Model

This study has chosen three types of discrete choice models to estimate labelling effects

and reference-dependent effects related to consumers‘ functional food choices: the Conditional

Logit (CL) model, the Random Parameter Logit (RPL) model and the Latent Class Model

(LCM). Although it has some limitations, the CL model is the standard starting point for deriving

other advanced discrete choice models. For those researchers who choose to use discrete choice

models, the CL model is a good first step. One major limitation of the CL model is that it

assumes that all respondents‘ preferences are homogeneous and the estimated coefficients are

fixed to be mean values of participants‘ responses.

To identify consumer preference heterogeneity, more advanced discrete choice models

need to be applied, such as the RPL model and LCM. For example, the results of the RPL model

indicate that strong variations exist in consumers‘ preferences for whether milk is enriched with

Omega-3. As suggested by McFadden and Train (2000), if a researcher‘s major concern is to

look for a highly flexible choice model, the RPL might be a good choice since it can approximate

any random utility model. LCM assumes that respondents can be intrinsically grouped into a

number of latent classes (Boxall and Adamowicz, 2002). Within each class, consumers‘

preferences are still assumed to be homogenous, but preferences are heterogeneous across

classes. For example, four distinctive latent classes have been identified in this study.

Furthermore, the class membership indicators in the LCM offer a better way to incorporate and

identify the sources of heterogeneity in consumer choices indicated in this study. Therefore, if

researchers believe that a discrete number of segments are sufficient to account for preference

heterogeneity, LCM is a good choice to capture the unobserved heterogeneity among classes.

172

Compared with the basic CL model, significant improvements in the goodness of fit were found

from the results of both the RPL model and the LCMs.

7.2 Implications

7.2.1 Implications for the Functional Food Industry

Health claims on food labels are one of the key issues in developing a successful market for

the functional foods industry, given the credence nature of functional attributes. This study

provides insights for the functional food sector through a better understanding of how consumers

perceive health claims and how labelling could influence their functional food choices. For

example, the results show that consumers respond positively to more detailed health claims (full

over partial labelling), and that having government or third party verifications could significantly

increase the acceptability of a new functional food compared with having no verification of

health claims. Although the use of a visual image (‗partial labelling‘) is the main labelling format

in the current Canadian functional foods market because of the particular labelling rules in

Canada, this study suggests that food manufacturers would benefit from the ability to make more

precise health claims. The results also suggest that the use of a heart symbol on functional food

products in the Canadian market may be just as effective as a function claim.

The CL model results indicate that, on average, consumers‘ levels of trust are close for a

government agency and a reputable third party for the verification of health claims. This result

shows that food manufacturers might have options when applying for certification of verified

health claims. The rules of obtaining verification of health claims might be more flexible from

third parties than from government agencies. However, clearly, the level of protection for

consumers might decrease if obtaining verification from third parties. Therefore, public policy

173

makers need to realize this type of possibility and develop appropriate policies to protect

consumers and maximize social welfare. Public policy implications are discussed in the next

section.

Moreover, the implications derived from the LCMs could help the Canadian functional

food industry to identify target consumer groups with different characteristics for the purpose of

developing marketing strategies. For example, the LCMs group the respondents into four latent

classes. The respondents in class 1 (―Price Sensitive Consumers‖, 55% of participants) appear to

be indifferent to functional milk with Omega-3. Price is the only determining factor when they

make milk purchase decisions. This result indicates that the majority consumers of milk products

still have no clear preference between conventional milk and Omega-3 enriched milk. Therefore,

reducing price could enable functional food companies to increase market share with this target

group. Class 2 (―Functional Milk Believers‖) and class 3 (―Verification Seekers‖) are the major

functional food consumers (about 40% of participants in total), whose purchase intention and

willingness to pay will increase with the presence of the three types of full health claims.

Therefore, food companies have strong incentives to lobby the regulatory agency governing

health claims to allow function claim or risk reduction claim for Omega-3 enhanced food

products.

The results also indicate that gender, income, health status, consumers‘ attitudes toward

functional foods, their health knowledge and their trust in health claims and nutrition labels are

sources of heterogeneity in consumers‘ preferences for functional foods. Therefore, functional

food manufacturers can focus their marketing efforts and target the consumer segments that are

more likely to purchase their products. For example, these results suggest that one target group

the Omega-3 milk companies could focus on is consumers who have heart disease, since those

174

consumers are more sensitive to the presence of Omega-3 in milk products and are willing to pay

more for Omega-3 milk. Functional food companies could also generally focus on higher income

consumers, since this study finds that they are willing to pay higher price premium for Omega-3

milk products. Of course, in the real word, food companies need to simultaneously consider

more influential factors beyond just income effect to expend their businesses. For example, the

result of this study shows that lower educated respondents with heart disease who have positive

attitudes toward functional food and higher level of trust are willing to pay the highest amount

for the risk reduction claim to be present on Omega-3 milk. The result also shows that female

consumers with higher income, positive attitudes towards functional food and lower health

knowledge are willing to pay the highest for milk enriched with Omega-3. These two groups of

consumers might be desirable target segments for the functional food manufacturers. This has

obvious implication for the types of messaging a functional food manufacturer would like to use

and the type of potential consumer these manufacturers will try to reach. However, these findings

should also meanwhile call attention of Canadian regulators to protect these groups of consumers

who might be particularly susceptible to misleading health claims.

It was also determined that professionals, such as doctors and nutritionists (94.3%), Health

organizations (90.1%) and Food labels (88%) are the top three health information sources that

respondents trust. This has implications for how to transmit information to potential consumers.

Food companies could inform these groups about the health benefits of Omega-3 milk which

might be an effective, albeit indirect, means of filtering this information through to final

consumers.

The loss aversion property from the reference-dependent model provides additional

insights into the functional food market. According to Prospect Theory, it is expected that

175

people will work harder to avoid losses than to obtain gains. Food companies might have an

incentive to emphasize ―disease (risk) reduction‖ rather than ―general health benefits‖, since

loss avoidances tend to have more influence on consumers‘ preferences than gains. Therefore,

functional food companies have incentives to use disease (risk) reduction functions on health

claims if they can successfully lobby for those types of disease (risk) reduction claims to be

used on food labels in Canada. This study also found that those respondents who already

frequently consume Omega-3 milk seem dislike to convert back to consume conventional milk.

Otherwise, their utility might significantly discount which is captured by the Omega-3 loss

effect. This is a clear and also important implication for the food industry to make effort to

provide more opportunities to consumers to consume Omega-3 milk. Price deduction or

promotion could be useful marketing strategies to have more consumers earn experience with

Omega-3 milk.

7.2.2 Public Policy Implications

In Canada, given the particular labelling environment regarding allowable health claims on

functional foods, a better understanding of consumers‘ reception to various health claims is

particularly important for making effective public policy choices. The results of this study

suggest that Canadian consumers are receptive to both full labelling (i.e. function claims, risk

reduction claims and disease prevention claims) and partial labelling (i.e. using symbol as cues)

formats. This is a clear indication that public policy makers need to pay attention to effectively

regulating health claims for functional foods. Overly restrictive regulation of health claims could

impair the communication between the functional food manufacturer and interested consumers

with respect to clear and valid health information by limiting the range and types of acceptable

health claims. However, overly lax regulation of health claims can leave consumers vulnerable to

176

misleading and deceptive health claims regarding apparent functional properties of foods. An

effective regulation system for health claims has the potential of promoting social welfare by

improving the health of Canadians and reducing health care costs. If the goal of public policy is

to identify which type of health claims could most effectively communicate health benefits to

consumers, the results in this study suggest that risk reduction claims might be the best option in

conveying health benefits to Canadian consumers and help them to make informed choices. Of

course, to be socially optimal, the regulatory system needs to establish standards to ensure the

risk reduction claims are truthful and not misleading, and based on sufficient health and

scientific evidence.

Another public policy implication of this study concerns the rules for function claims and

disease prevention claims. In Canada, the rules surrounding the use of function claims may be

more onerous than that in the United States. However, this type of ‗restrictive‘ rule might be

necessary in terms of protecting consumers from misleading health claims. As the analysis in this

study shows the use of a heart symbol, as currently used in the Canadian functional foods market,

may be just as effective in communicating a health benefit as using a function claim. Therefore,

it might be necessary for the development of a set of standards and rules to regulate the food

industry using a symbol or visual cue on functional food from the point of view of consumer

protection. Currently, disease prevention claims are regulated as drug claims and not permitted to

be used on food products in both Canada and the United States. The results in this study indicate

that credible disease prevention claims are not resisted by many Canadian consumers for

functional food products. Therefore, if enough scientific evidence became available to support

the role of functional foods in preventing certain types of chronic diseases, policy makers may

want to consider the option of allowing certain disease prevention claims to be applied to food

177

products in the future. Again, from the perspective of consumer protection it will be critical that

sufficient scientific evidence exists to warrant such a strong health claim. If disease prevention

claims were allowed, consumers are very likely to believe and respond to them, so tight

regulation of allowable claims remains very important in protecting consumers.

If the regulatory system is equivocal on what types of health claims should be approved

and how those health claims are approved, it will be difficult for the system to protect consumers

from misleading health claims and also limit the development of the Canadian functional food

industry. Indeed, it remains an important challenge for regulators to develop an efficient system

that balances effectively communicating health information and preventing deceptive claims, and

balances the imperative of protecting consumers‘ interests and satisfying companies‘ marketing

requirements.

Claims about health effects are credence attributes, which means even in the presence of a

labelled health claim, information asymmetry may still be present if consumers are uncertain

about the validity of the health claim. Public policy makers should be aware that the verification

of health claims plays an important role in reducing consumers‘ uncertainty and making health

claims more credible. The quality assurance made by verification organizations provides a means

of establishing the credibility of health claims. The results in this study suggest that Canadian

consumers are more likely to choose functional food products if a government agency (e.g.

Health Canada) or a reputable third party (e.g. Heart and Stroke Foundation) has verified the

health claims. Therefore, it is also important for the government regulatory agency to codify and

standardize the rules for third party endorsement of health claims. A set of accurate and

transparent standards are very important not only for regulating truthful and not misleading

health claims, but also for monitoring the verification of health claims by different organizations.

178

7.3 Limitations and Future Research

This thesis focuses on examining the labelling and reference-dependent effects in Canadian

consumers‘ functional food choices. A choice experiment is adopted as the research

methodology. One limitation of this study is associated with the hypothetical nature of the stated

preference approach, since respondents are asked to state their preference values but actual

choice behaviour may differ. Consumers may provide unrealistic statements if there is no cost to

over (or under-) stating their willingness to pay. Estimation bias may be present due to strategic

behaviour by respondents, especially when consumers are unfamiliar with the product (e.g. a

food product with a new functional attribute), their stated willingness to pay may be inaccurate.

In this study, although Omega-3 milk products have already been available in the Canadian

functional food market for quite a few years, health claims pertaining to Omega-3 are still not

allowed in Canada. Therefore, the examined product, Omega-3 milk with potential health claims,

is still a hypothetical product. Other methods or the revealed preference method could be applied

if health claims for Omega-3 are allowed in the Canadian functional food market in the future.

Other methods, such as experimental auctions, have been widely discussed in the literature

associated with economic evaluation methods and could be used in future research on this topic.

As discussed by Hu, Adamowicz and Veeman (2006), it is broadly believed that the use of

experimental auctions in consumer research can capture the true willingness to accept a product

and reduce the bias caused by strategic behaviour. However, the costs of conducting auctions on

representative samples are usually relatively higher than the Stated Preference Method and as a

result, sample sizes tend to be smaller. The other drawback of auctions that precluded their use in

this study is the requirement to have the product (in this case Omega-3 milk with different

labelled health claims) available for respondents to ‗purchase‘ in the auction. Nevertheless,

179

comparison of the methods and results between those two types of data (the Stated Preference

Method vs. auction) might be interesting for future research.

Data for this study were collected through an online survey across Canada with the

exception of Quebec in the summer of 2009. Thus, another limitation of this study is that the

survey data omits respondents from Quebec since the survey was conducted only in English due

to resource constraints. Future research could expand to include respondents in Quebec, where

food choices and attitudes towards health may differ for cultural reasons.

Furthermore, the effects of full labelling and partial labelling are only derived from a

single product, Omega-3 milk. Two additional caveats are particularly important. First, the

analysis only examines heart health claims and the use of a heart symbol for Omega-3 milk, and

the connection between the two labelling formats is fairly self-evident. The claims pertaining to

other health conditions (such as digestive health) may be more difficult to convey clearly through

the use of visual imagery. Furthermore, it is unclear whether there are ―spill-over‖ effects

between different categories of functional food products. For example, whether consumers‘

attitudes and purchase behaviour for Omega-3 milk in this study could be used to explain

consumers‘ acceptance of other categories of functional foods, such as vitamin enriched juice or

mineral enriched cereals. Second, the issue of consumer protection from fraudulent or misleading

health claims has not been addressed in this study. Future research could focus on examining the

existence of spill-over effects and the issues of protecting consumers from misleading health

claims.

Considering the emergent nature of the functional food industry in Canada, future studies

should take a longitudinal approach in tracking the dynamics and co-evolution among

180

government policies, industry practices, and consumer behaviours and health outcomes. A

related future research area is the economic analysis of the labelling effects and verification

effects along the supply chain of functional food products, including the analyses of strategic

decisions by farmers/producers, processors, retailers, exporters and other stakeholders. For

example, the regulatory uncertainty and the credibility problem for health claims could influence

producers‘ investment in research and development for novel functional foods. It could directly

affect the differentiation and availability of functional food products produced by processors and

marketed by retailers. Therefore, this is a rich area for further research analysis.

One more extension of this research is related to modelling reference point change effects

and making comparisons of respondents‘ attitudes over multiple periods of time. In this study,

the data were collected within one survey and respondents‘ reference points are assumed to be

fixed. Over time, respondents‘ reference levels might change with their familiarity with the

examined products in the market which may reshape their decision making processes. For

example, if risk reduction health claims are permitted for Omega-3 milk in the future, it would be

interesting to see whether individual‘s preferences change as a result of changes in their

reference points for health claims. This type of study would need consumer panel data where

consumers‘ preferences for product attributes could be tracked by researchers over time. The use

of longitudinal consumer panels is a promising avenue for future research.

181

References:

Agriculture and Agri-Food Canada (AAFC) (1995). Nutraceuticals/Functional Foods

Exploratory Survey on Canada‘s Potential. Monograph, 1995, Canada.

Agriculture and Agri-Food Canada (AAFC), News Release (2007). Canada‘s new government

invests $721,000 in natural health products. Available online at:

http://www.agr.gc.ca/cb/index_e.php?s1=n&s2=2007&page=n70605 (Accessed on August 9,

2010)

Agriculture and Agri-Food Canada (2009). Functional Foods and Nutraceuticals. Commentaries.

http://www4.agr.gc.ca/AAFC-AAC/display-afficher.do?id=1170856376710 (accessed on Nov

15, 2010)

Adamowicz, W. L., J. J. Louviere and M. Williams (1994). Combining Revealed and Stated

Preference Method for Valuing Environmental Amenities. Journal of Environmental Economics

and Management, 26: 271-292.

Adamowicz, W. L., J. D. Swait, P. C. Boxall, J. J. Louviere and M. Williams (1997). Perceptions

Versus Objective Measures of Environmental Quality in Combined Revealed and Stated

Preference Models of Environmental Valuation. Journal of Environmental Economics and

Management, 32: 65-84.

Adamowicz, W. L., P. Boxall, M. Williams and J. Louviere (1998). Stated Preference

Approaches for Measuring Passive Use Values: Choice Experiments and Contingent Valuation.

American Journal of Agricultural Economics, 80(February): 64-75.

Akerlof, G. A. (1970). The Market for ‗Lemons‘: Quality Uncertainty and the Market

Mechanism. Quarterly Journal of Economics, 84: 488-500.

Anders, S. and A. Moser (2010). Consumer Choice and Health: the Importance of Health

Attributes for Retail Meat Demand in Canada. Canadian Journal of Agricultural Economics, 58

(2): 249-271.

Barry, B. and T. S. Bateman (1996). A Social Trap Analysis of the Management of Diversity.

The Academy of Management Review, 21(3): 757-790.

Basu, K. S., J. E. Thomas and S. N. Acharya (2007). Prospects for Growth in Global

Nutraceutical and Functional Food Markets: A Canadian Perspective. Australian Journal of Basic

and Applied Sciences, 1(4): 637-649.

Batsell, R. R. and J. J. Louviere (1991) Experimental Analysis of Choice. Marketing Letters,

2(3): 199-214.

182

Bech-Larsen, T. and K. G. Grunert (2003). The Perceived Healthiness of Functional Foods: a

Conjoint Study of Danish, Finnish and American Consumers‘ Perception of Functional Foods.

Appetite, 40(1): 9-14.

Bell R. D. and J. M. Lattin (2000). Looking for Loss Aversion in Scanner Panel Data: The

Confounding Effect of Price Response Heterogeneity. Marketing Science, 19(2): 185-200.

Ben-Akiva, M. and S.R. Lerman. (1985). Discrete Choice Analysis : Theory and Application to

Travel Demand. Cambridge, Mass: MIT Press.

Bhat, C. (1998). Accommodating Variations in Responsiveness to Level-of-Service Measures in

Travel Mode Choice Modeling. Transportation Research Part A, (32): 455-507.

Blamey, R., J. J. Louviere and J. Bennett (2001). ‗Choice Set Design‘ in the Choice Modeling

Approach to Environmental Valuation. Ed. J. Bennett an R. Blamey. Northampton. Publisher:

Edward Elgar.

Blend, J. R. and E. O. van Ravenswaay. (1999). Measuring consumer demand for ecolabeled

apples. American Journal of Agricultural Economics, 81: 1072–1077.

Bloch A., and C A. Thomson. (1995). Position of the American Dietetic Association:

Phytochemicals and Functional Foods. Journal of the American Dietetic Association, ADA

report, 95(4): 493-496.

Bond C. A., D. D. Thilmany and J. K. Bond (2008). What to Choose? The Vaule of Label

Claims to Fresh Produce Consumers. Journal of Agricultural and Resource Economics, 33(3):

402-427.

Boxall, C. P. and W. L. Adamowicz (2002). Understanding Heterogeneous Preferences in

Random Utility Models: A Latent Class Approach. Environmental and Resource Economics, 23:

421-446.

Boyd, J. and J. Mellman (1980). The Effect of Fuel Economy Standards on the U.S. Automotive

market: A Hedonic Demand Analysis. Transportation Research A 14, 367-378.

Briesch, R.A., L. Krishnamurthi, T. Mazumdar, and S.P Raj. (1997). A Comparative Analysis of

Reference Price Models. Journal of Consumer Research, 24(2): 202-214.

Burton, M., D. Rigby, T. Young and S. James (2001). Consumer Attitudes to Genetically

Modified Organisms in Food in the UK. European Review of Agricultural Economics, 28 (4):

479-498.

183

Canadian Food Inspection Agency (2009). Guide to Food Labelling and Advertising, Chapter 8-

Health Claims. Regulatory Document.

http://www.inspection.gc.ca/english/fssa/labeti/guide/ch8e.shtml (accessed on October 1, 2010)

Carson, R. T., J. J. Louviere, D. A. Anderson, P. Arabie, D. S. Bunch, D. A. Hensher, R. M.

Johnson, W. F. Kuhfeld, D. Steinberg, J. Swait, H. Timmermans and J. B. Wiley (1994).

Experimental Analysis of Choice. Marketing Letters, 5(4), 351-368.

Cash, S. B., E. W. Goddard and M. Lerohl (2006). Canadian Health and Food: the Links

between Policy, Consumers and Industry. Canadian Journal of Agricultural Economics, 54(4):

605-629.

Caswell, J. A. (2000). An evaluation of risk analysis as applied to agricultural biotechnology.

Agribusiness, 16(1): 115–23.

Caswell, J., C. Noelke and E. Mojduszka (2002). Unifying Two Frameworks for Analyzing

Quality and Quality Assurance for Food Products. In Global Food Trade and Consumer Demand

for Quality, edited by B. Krissoff, M. Bohman and J. Caswell, 43-61. New York: Kluwer

Academic Publishers.

CFIA [Canadian Food Inspection Agency] (2001). Labelling of Foods from Biotechnology in

Canada. http://www.inspection.gc.ca/english/sci/biotech/conse.shtml#labeti

Chase, D., T. Nilsson, J. P. Emunu, D. McCann-Hiltz and Y. Peng (2007). Canadian Consumer

Attitudes and Purchasing Behaviour of Omega-3 Products. Working paper, Research Project

Number CMD-07-04, Consumer & Market Demand Agricultural Policy Research Network.

Cox, D.N., A. Koster and C. G. Russell (2004). Predicting Intentions to Consumer Functional

Foods and Supplements to Offset Memory Loss Using an Adaptation of Protection

Motivation Theory. Appetite, 43: 55-64.

Cranfield, J., B. J. Deaton and S. Shellikeri (2009). Evaluating Consumer Preferences for

Organic Food Production Standards. Canadian Journal of Agricultural Economics, 57(1): 99-117.

Cronbach, L. J. (1951). Coefficient Alpha and the Internal Structure of Test. Psychometrika,

16(3): 297- 334.

Crowley, L. (2008). Canada seeks opinion on health claims review. News Release. Available

online at: http://www.nutraingredients.com/Product-Categories/Vitamins-premixes/Canada-

seeks-opinion-on-health-claims-review (accessed on July 7, 2010)

CSIRO Human Nutrition (2004). Functional Foods. CSIRO Human Nutrition, Australia. http:

//www.foodscience.csiro.au/functional-foods.htm (accessed May 2006).

184

Darby, M.E. and E. Karni (1973). Free Competition and the Optimal Amount of Fraud. Journal

of Law and Economics, 16(1): 67-88.

DeFelice, S. (2007). DSHEA Versus NREA (The Nutraceutical Research and Education Act)

and the Three Nutraceutical Objectives. The Foundation for Innovation of Science in Medicine.

Commentaries.

DeJong, N., M.C. Ocke, H.A.C. Branderhorst and R. Friele (2003). Demographic and

Lifestyle Characteristics of Functional Food Consumers and Dietary Supplement Users. British

Journal of Nutrition, 89(2): 273-81.

DeLorme, D., J. Huh, and L. N. Reid (2009). DTC Advertising Skepticism and

the Use and Perceived Usefulness of Prescription Drug Information Sources. Health Marketing

Quarterly, 26: 293–314.

Dhami, S. and al-Nowaihi, A. (2007). Why do people pay taxes? Prospect theory Versus

Expected Utility Theory. Journal of Economic Behaviour & Organization, 64: 171-192.

D‘Innocenzo, L. (2006). The future is ripe. Bio Business, XII: 20-24.

Doyon, M. and J. Labrecque (2008). Functional Food: a Conceptual Definition. British Food

Journal, 110(11): 1133-1149.

Drazen, P. (1998). The Probability Weighting Function. Econometrica, 66(3): 497-527.

Foreign Affairs and International Trade Canada (FAIT) (1995). Canada-Japan Collaborative

Research and Development for Food Ingredients, Additives and Functional Foods: Identifying

Potential Canadian Partners. Monograph, 1995, Canada.

FAO/WHO (2001). Evaluation of health and nutritional properties of powder milk and live lactic

acid bacteria. Food and Agriculture Organization of the United Nations and World Health

Organization Expert Consultation Report.

FDA (2004). FDA announced qualified health claims for omega-3 fatty acids.

http://www.fda.gov/SiteIndex/ucm108351.htm (accessed on July 8, 2010)

FDA (2009a). 8: Claims. Guidance for Industry: A Food Labelling Guide. FDA Guidance

Document.

http://www.fda.gov/Food/GuidanceComplianceRegulatoryInformation/GuidanceDocuments/F

oodLabelingNutrition/FoodLabelingGuide/ucm064908.htm#health (accessed on March 1, 2011)

FDA (2009b). Appendix C: Health Claims. Guidance for Industry: A Food Labelling Guide. FDA

Guidance Document.

http://www.fda.gov/Food/GuidanceComplianceRegulatoryInformation/GuidanceDocuments/F

oodLabelingNutrition/FoodLabelingGuide/ucm064919.htm (accessed on March 1, 2011)

185

FDA (2009c). Appendix D: Qualified Health Claims. Guidance for Industry: A Food Labelling

Guide. FDA Guidance Document.

http://www.fda.gov/Food/GuidanceComplianceRegulatoryInformation/GuidanceDocuments/Foo

dLabelingNutrition/FoodLabelingGuide/ucm064923.htm (accessed on March 1, 2011)

Federal Trade Commission (2006). Before the Department of Health and Human Services, Food

and Drug Administration. In the Matter of Assessing Consumer Perceptions of Health

Claims; Public Meeting; Request for Comments, Docket No. 2005N-0413. January 17.

Fitzpatrick, C. K. (2003). Functional foods and nutraceuticals: exploring the Canadian

Possibilities. Research Report, University of Manitoba, Winnipeg.

Fitzpatrick, C. K. (2004). Regulatory Issue Related to Functional Foods and Natural Health

Products in Canada: Possible Implications for Manufacturers of Conjugated Linoleic Acid. The

American Journal of Clinical Nutrition, 79: 1217S-1020S.

Fowler, F. J. (1995). Improving Survey Questions: Design and Evaluation. Applied Social

Research Methods Series, Volume 38. Sage Publication, Inc. Library of Congress Cataloging-in-

Publication Data. Printed in the United States of America

Galesic, M. and M. Bosnjak (2009). Effects of Questionnaire Length on Participation and

Indicators of Response Quality in a Web Survey. Public Opinion Quarterly, 73(2): 349-360.

Garretson, J.A. and S. Burton (2000). Effects of Nutrition Facts Panel Values, Nutrition Claims,

and Health Claims on Consumer Attitudes, Perceptions of Disease-Related Risks, and

Trust. Journal of Public Policy & Marketing, 19(2): 213-227.

Gilbert, L. (2000). Marketing Functional Foods: How to Reach Your Target Audience.

AgBioForum, 3(1): 20-38.

Greene, P. E. and V. Srinivasan (1978). Conjoint Analysis in Consumer Research: Issues and

Outlook. Journal of Consumer Research, 5: 103-123

Hailu, G., A. Boecker, S. Henson and J. Cranfield (2009). Consumer valuation of functional

foods and nutraceuticals in Canada. A conjoint study using probiotics. Appetite, 52:

257-265.

Hair, F. J., R. E. Anderson, R. L. Tatham and W. C. Black (1992). Multivariate Data Analysis

with Readings. 3rd

Edition. Macmillan Publishing Company, New York.

Hanemann W. M. (1994). Valuing the Environment Through Contingent Valuation. The Journal

of Economic Perspectives, 8(4): 19-43.

Hardie, B. G. S., E. J. Johnson and P. S. Fader (1993). Modelling Loss Aversion and Reference

Dependence Effects on Brand Choice. Marketing Science, 12(4): 378-394.

186

Health Canada (1998a). Nutraceutical and Functional Food and Health Claims on Foods. Health

Canada, Policy Paper. Available online at: http://www.hc-sc.gc.ca/fn-an/label-etiquet/claims-

reclam/nutra-funct_foods-nutra-fonct_aliment-eng.php (Accessed on August 18, 2010)

Health Canada, (1998b). Regulations Amending the Food and Drug Regulations (948 — Novel

Foods). Food and Drugs Act, Canada Gazette, part II, 133.22: 2376 – 2439.

http://canadagazette.gc.ca/partII/1999/19991027/pdf/g2-13322.pdf (Accessed on August 18,

2010)

Health Canada (2003). Regulations Amending the Food and Drug Regulations (Nutrition

Labeling, Nutrition Claims and Health Claims). Canada Gazette Part II. January 2003.

http://gazette.gc.ca/archives/p2/2003/2003-01-01/html/sor-dors11-eng.html (Accessed on July 1,

2010)

Health Canada (2007). Managing Health Claims for Foods in Canada: Towards a Modernized

Framework. Discussion Paper, Food Directorate, Health Products and Food Branch, Health

Canada. http://www.hc-sc.gc.ca/fn-an/consult/_man-gest_health_claims-allegations_sante/index-

eng.php (Accessed on September 1, 2009)

Health Canada (2010a). Food and Nutrition: Health Claims Assessments. http://www.hc-

sc.gc.ca/fn-an/label-etiquet/claims-reclam/assess-evalu/index-eng.php (accessed on January 15,

2010)

Health Canada (2010b). Food and Nutrition: Oat products and Blood Cholesterol Lowering.

http://www.hc-sc.gc.ca/fn-an/label-etiquet/claims-reclam/assess-evalu/oat-avoine-eng.php

(accessed on January 15, 2010)

Health Canada (2010c). Food and Nutrition: Plant Sterols (Phytosterols) in Food. http://www.hc-

sc.gc.ca/fn-an/label-etiquet/claims-reclam/assess-evalu/phytos-index-eng.php (accessed on

January 15, 2010)

Heller I. R., Y Taniguchi and T. Lobstein (1999). Functional Foods: Public Health Boon or 21st

Century Quackery? the International Association of Consumer Food Organizations Report in

1999. http://www.cspinet.org/reports/functional_foods/index.html (accessed on July 8, 2010)

Hensher, D. A., J. M. Rose, and W. H. Greene (2005). Applied Choice Analysis, A Primer.

Cambridge Press, New York.

Hilliam, M. (2000). Functional food-How big is the market? The world of Food Ingredients, 12:

50-52.

Holmes, T. P. and W. A. Adamowicz (2003). Chapter 6: Attribute-Based Methods. in A Primer

on Nonmarket Valuation. edited by P.A. Champ K.J. Boyle and T.C. Brown. Dordrecht: Kluwer

Academic Publishers, 171-219.

187

Hovde C. S., C. J. Wachenheim, R. Hearne and W. Nganje (2007). Identifying Market

Preferences for High Selenium Beef. Working paper, Agribusiness and Applied Economics

Report No. 611. Availabe online at: http://ageconsearch.umn.edu/bitstream/7633/1/aer611.pdf

(Accessed on August 17, 2010)

Hu, W., M. M. Veeman and W. L. Adamowicz (2005). Labelling Genetically Modified Food:

Heterogeneous Consumer Preferences and the Value of Information. Canadian Journal of

Agricultural Economics, 53(1): 83-102.

Hu, W., W. L. Adamowicz. and M. M. Veeman (2006). Labelling Context and Reference Point

Effects in Models of Food Attribute Demand. American Journal of Agricultural Economics,

88(4): 1034-1049.

Hu, W., T. Woods and S., Bastin (2009). Consumer Acceptance and Willingness to Pay for

Blueberry Products with Nonconventional Attributes. Journal of Agricultural and Applied

Economics, 41(1): 47-60.

Huber, J. and K. Zwerina (1996) The Importance of Utility Balance and Efficient Choice

Designs, Journal of Marketing Research, 33: 307-317.

Innes, B.G. and J.E. Hobbs (2011). Does it Matter Who Verifies Production-Derived Quality?

Canadian Journal of Agricultural Economics, 59(1): 87-107.

Jain, D., N. Vilcassim, and P. Chintagunta (1994). A Random-Coefficients Logit Brand Choice

Model Applied to Panel Data. Journal of Business and Economic Statistic, 12(3): 317-328.

Jallinoja, P., and A. R. Aro (2000). Does Knowledge Make a Difference? The Association

Between Knowledge About Genes and Attitudes Toward Gene Tests. Journal of Health

Communication, 5(1): 29 - 39.

Johnson, F. R., B. Kanninen, M. Bingham and S. Ozdemir (2007). Experimental Design for

Stated Choice Studies, in Valuing Environmental Amenities Using Stated Choice Studies. In B. J.

Kanninen (ed.), 159-202. Springer, Dordrecht, the Netherlands.

Kahneman, D. and A. Tversky (1979). Prospect Theory: An Analysis of Decision under Risk,

Econometrica, 47(2): 263-292.

Kahneman, D., J. L. Knetsch and R. Thaler (1990). Experimental Tests of the Endowment Effect

and the Coase Theorem. Journal of Political Economy, XCVIII, 1325-1348.

Kalwani, M. U., C. K. Yim, H. J. Rinne and Y. Sugita (1990). A Price Expectations Model of

Customer Brand Choice. Journal of Marketing Research, 27(August): 251-262.

Kalwani, M. U. and C. K. Yim (1992). Consumer Price and Promotion Expectations: An

Experimental Study. Journal of Marketing Research, 29(February): 90-100.

188

Kanninen, B. J. (2002). Optimal Design for Multinomial Choice Experiments, Journal of

Marketing Research, 39: 214-217.

Kleef, E., H. C.M. Trijp and P. Luning (2005). Health claim-food product compatibility and the

impact of health claim framing on consumer evaluation, Appetite, 44(2005): 200-308

Koszegi, B. and M. Rabin (2006). A Model of Reference-Dependent Preferences, Quarterly

Journal of Economics, 121: 1133-1166.

Kozup, J.C., E.H. Creyer, and S. Burton (2003). Making Healthful Food Choices: The Influence

of Health Claims and Nutrition Information on Consumers‘ Evaluations of Packaged

Food Products and Restaurant Menu Items. Journal of Marketing, 97(April): 19-34.

Kroes, P. E. and R. J. Sheldon (1988). Stated Preference Methods: An Introduction. Journal of

Transport Economics and Policy, 22(1): 11-25.

Krosnick, J. A. and D. F. Alwin (1987). An Evaluation of a Cognitive Theory of Response Order

Effects in Survey Measurement. Public Opinion Quarterly, 51: 201-219.

Krosnick, J. A. (1991). Response Strategies for Coping with the Cognitive Demands of Attitude

Measure in Surveys. Applied Cognitive Psychology, 5: 213-236.

Kuhfeld, W. F., R. D. Tobias, and M. Garratt (1994) Efficient Experimental Design with

Marketing Research Applications. Journal of Marketing Research, 21: 545-557.

Kuhfeld, W. F. (1997). Efficient Experimental Designs Using Computerized Searches. Working

paper for the Sawtooth Software Research Paper Series, SAS Institute, Inc. Available online at:

http://www.cs.uiowa.edu/~gwoodwor/AdvancedDesign/KuhfeldTobiasGarratt.pdf (Accessed on

August 10, 2010)

Kuhfeld, W. F. (2005) Marketing Research Methods in SAS: Experimental Design, Choice,

Conjoint, and Graphical Techniques. SAS 9.1 Edition, TS-722, Cary, NC: SAS Institute.

L‘Abbe, R. M., L. Dumais, E. Chao and B. Junkins (2008). Health Claims on Foods in Canada.

The Journal of Nutrition, 138: 1221S-1227S.

Labreque, J., M. Doyon, F. Bellavance and M. Kolodinsky (2006). Acceptance of Functional

Foods: A Comparison of French, American, and French Canadian Consumers. Canadian

Journal of Agricultural Economics, 54: 647-661.

Laibson, D. and R. Zeckhauser (1998). Amos Tversky and Ascent of Behavioural Economics.

Journal of Risk and Uncertainty, 16: 7-47.

Lancaster, K. J. (1966). A new approach to consumer theory. Journal of Political Economy, 74:

132-157.

189

Larue, B., G. E. West, C. Gendron and R. Lambert (2004). Consumer Response to Functional

Foods Produced by Conventional, Organic, or Genetic Manipulation. Agribusiness, 20(2): 155-

166, published online in Wiley InterScience: www.interscience.wiley.com

Lattin, J. M. and R. E. Bucklin (1989). Reference Effects of Price and Promotion on Brand

Behaviour. Journal of Marketing Research, 26: 299-310.

Lindlof, T. R. and B. C. Taylor (2002). Qualitative Communication Research Methods, 2nd

Edition. Thousand Oaks, CA: Sage.

Louviere, J. J. (1988). Conjoint Analysis Modeling of States Preference: A Review of Theory,

Method, Recent Developments and External Validity. Journal of Transport Economics and

Policy, 22: 93-119.

Louviere, J. J., D. A. Hensher and J. D. Swait (2000). Stated Choice Methods, Analysis and

Application. Cambridge Press, United Kingdom.

Louviere, J. J. and G. Woodworth (1983). Design and Analysis of Simulated Consumer Choice

or Allocation Experiments: An Approach Based on Aggregate Data. Journal of Marketing

Research, 20: 350-367.

Lusk, L. J. and N. Parker (2009). Consumer Preferences for Amount and Type of Fat in Ground

Beef. Journal of Agricultural and Applied Economics, 41(1): 75-90.

Malhotra, N. (2008). Completion Time and Response Order Effects in Web Surveys. Public

Opinion Quarterly, 72(5): 914-934.

Malla, S., J. E. Hobbs and O. Perger (2007). Valuing the Health Benefits of a Novel Functional

Food. Canadian Journal of Agricultural Economics, 55: 115-136.

Malla, S.M., J.E. Hobbs and O. Perger (2005). Estimating the Health Care Savings from Trans

Fat-Free Canola in Canada. Food Economics, 2(3-4): 175-184.

Manski, C. (1977). The Structure of Random Utility Models. Theory and Decision, 8: 229-254.

McCutcheon, A. L. (1987). Latent Class Analysis. Sage University Papers Series: Quantitative

Applications in the Social Sciences Number 07-064. Newbury Park, CA: Sage Publications.

McFadden, D (1974). Conditional Logit Analysis of Qualitative Choice Behavior. In Frontiers in

Econometrics. P. Zarembka (ed). New York: Academic Press. Chapter four: 105-142. Available

online at: http://elsa.berkeley.edu/reprints/mcfadden/zarembka.pdf (Accessed on August 10,

2010)

McFadden, D (1986). The Choice Theory Approach to Market Research, Marketing Science,

5(4): 275-297.

McFadden, D. and K. Train (2000). Mixed MNL Models of Discrete Response. Journal of

Applied Econometrics, 15: 447-470.

190

Menrad, M., B. Husing, K. Menrad, T. Rei, S. Beer-Borst, and C. A. Zenger (2000). Functional

Food. TA 37/2000. Bern: Schweizerischer Wissenschafts und Technologierat.

Menrad K. (2003). Market and marketing of functional food in Europe, Journal of Food

Engineering, 56: 181–188.

Miljkovic, D. (2005). Rational Choice and Irrational Individuals or Simply an Irrational Theory:

A Critical Review of the Hypothesis of Perfect Rationality. The Journal of Socio-Economics,

34(5): 621-634.

Morikawa, T., M. Ben-Akiva and D. McFadden (2002). Discrete Choice Models Incorporating

Revealed Preferences and Psychometric Data. Econometric Models in Marketing, 16: 29-55.

Munro, A. and R. Sugden (2003). On the theory of Reference-Dependent Preferences. Journal

of Economic Behaviour & Organization, 50: 407-428.

Nelson, P. (1970). Information and Consumer Behaviour. Journal of Political Economy, 81:

729-754.

Nilsson, T., K. Foster and J. L. Lusk (2006). Marketing Opportunities for Certified Pork Chops.

Canadian Journal of Agricultural Economics, 54: 567-583.

Nutrition Business Journal (2007). The Global Nutrition Industry VI. Nutrition Business Journal,

5/6: 3–47.

Ordonez, L. D. (1998) The Effect of Correlation Between Price and Quality in Consumer

Choice. Organizational Behavior and Human Decision Processes, 75(3): 258-273.

Peng, Y., G. E. West and C. Wang (2006). Consumer Attitudes and Acceptance of CLA-

Enriched Dairy Products. Canadian Journal of Agricultural Economics 54(4): 663-684

Poulsen, J. (1999). Danish consumers‘ attitudes towards functional foods. MAPP working

paper, 62. Aarhus School of Business. www.mapp.asb.dk/wppdf/wp62.pdf

Prelec, D. (1998). The Probability Weighting Function. Econometrica, 66(3): 497-527.

Putler, D. S. (1992). Incorporating Reference Price Effects into a Theory of Consumer Choice.

Marketing Science, 11 (Summer): 287-309.

Quagrainie, K. K., J. Unterschultz and M. Veeman (1998). Effects of Product Origin and

Selected Demographics on Consumer Choice of Red Meats. Canadian Journal of Agricultural

Economics, 46: 201-219.

191

Reid, G., M. E. Sanders, H. R. Gaskins, G. R. Gibson, A. Mercenier, R. Rastall, M. Roberfroid, I.

Rowland, C. Cherbut, R.T. Klaenhammer (2003). New Scientific Paradigms for Probiotics and

Prebiotics. Journal of Clinical Gastroenterology, 37(2): 105–118.

Roe, B., A.S. Levy, and B.M. Derby (1999). The Impact of Health Claims on Consumer Search

and Product Evaluation Outcomes: Results from FDA Experimental Data. Journal of

Public Policy & Marketing, 18(1): 89-105.

Roosen, J., J. L. Lusk and J. A. Fox (2003). Consumer demand for and attitudes toward

alternative beef labelling strategies in France, Germany, and the UK. Agribusiness, 19(1): 77–

90.

Rousu, M., W. Huffman, J. F. Shogren and A. Tegene (2004). Estimating the public value of

conflicting information: The case of genetically modified foods. Land Economics, 80(1): 125–

35.

Samuelson, P. (1938). A Note on the Pure Theory of Consumers' Behaviour. Economica, 5: 61-

71.

Shen, J. (2009) Latent class model or mixed logit model? A comparison by transport mode

choice data. Applied Economics, 41(22): 2915 – 2924.

Smith, B. L., M. Marcotte and G., Harrison (1996). A Comparative Analysis of the Regulatory

Framework Affecting Functional Food Development and Commercialization in Canada, Japan,

the European Union and the United States of America, Monograph, March 31, Canada.

Sorenson, D. and J. Bogue (2005). Market-Oriented New Product Design of Functional

Orange Juice Beverages: A Qualitative Approach. Journal of Food Products Marketing, 11(1):

57-73.

Statistics Canada (2003). Functional Foods and Nutraceutical Survey 2003. Available online at:

http://www.statcan.ca/Daily/English/031006/d031006c.htm (Accessed on August 16, 2010).

Statistics Canada (2007). Functional Foods IX. Functional Foods and Nutraceuticals Survey.

Nutrition Business Journal. http://investincanada.gc.ca/eng/industry-sectors/agri-food/natural-

health.aspx (accessed August 16, 2010)

Statistics Canada (2008). Census 2006. Available online at: http://www12.statcan.ca/census-

recensement/2006/as-sa/index-eng.cfm (Accessed on August 20, 2010).

Steiner, B., F. Gao and J. Unterschultz (2010). Alberta Consumers‘ Valuation of Extrinsic and

Intrinsic Red Meat Attributes: A Choice Experimental Approach. Canadian Journal of

Agricultural Economics, 58(2): 171-189.

192

Suzuki, Y., J. E. Tyworth and R. A. Novack (2001). Airline Market Share and Customer Service

Quality: A Reference-Dependent Model. Transportation Research Part A, 35: 773-788.

Tebbens, J. (2005). Functional foods and nutraceuticals: the development of value-added food by

Canadian firms. Statistics Canada. Science, Innovation and Electronic Information Division.

Available online at : http://dsp-

psd.pwgsc.gc.ca/Collection/Statcan/88F0006X/88F0006XIE2005016.pdf (Accessed on August

10, 2010)

Train, K. (1998). Recreation Demand Models with Taste Variation. Land Economics, 74: 230-

239.

Train K. E. (2003). Discrete Choice Methods with Simulation, Cambridge University Press.

Tversky, A. and D. Kahneman (1981). The Framing of Decisions and the Psychology of Choice.

Science, New Series, 211(4481): 453-458.

Tversky, A. and D. Kahneman (1991). Loss Aversion in Riskless Choice: A Reference-

Dependent Model. The Quarterly Journal of Economics, 106(4): 1039-1061.

Uzea, A.D., J.E. Hobbs and J. Zhang. (2010). Activists and Animal Welfare: Quality

Verifications in the Pork Sector. Working paper. Department of Bioresource Policy, Business &

Economics, University of Saskatchewan

Verbeke, W. (2005). Consumer Acceptance of Functional Foods: Socio-Demographic, Cognitive

and Attitudinal Determinants. Food Quality and Preference, 16: 45-57.

Wansink, B. (2003). How Do Front and Back Package Labels influence Beliefs about Health

Claims? The Journal of Consumer Affairs, 37(2): 305-316.

West, G. E., C. Gendron, B. Larue and R. Lambert (2002). Consumers‘ Valuation of Functional

Properties of Foods: Results from a Canada-wide Survey. Canadian Journal of Agricultural

Economics, 50(4): 541-558.

Weststrate, J.A., van Poppel, G. and Verschuren, P.M. (2002). Functional foods, trends and

Future. British Journal of Nutrition, 88(S2): 233-235.

Winer, R. S. (1986). A Reference Price Model of Brand Choice for Frequently Purchased

Products. Journal of Consumer Research, 13(September): 250-256.

Winer, R. S. (1988). Behavioral Perspective on Pricing: Buyers' Subjective Perceptions of Price

Revisited, in T. M. Devinney (Ed.), Issues in Pricing. Theory and Research, Lexington, MA:

Lexington Books.

World Health Organization (WHO) (2002). World Health Report. World Health Organization,

Geneva.

193

World Health Organization (WHO) (2005). Preventing Chronic Diseases: a Vital Investment.

http://www.who.int/chp/chronic_disease_report/en/ (accessed on August 15, 2010)

World Nutraceuticals (2006). Industry Study with Forecasts to 2010 & 2015. The Freedonia

Group, Cleveland, OH USA.

Zou, N., and J. Hobbs (2006). Modeling Functional Food Choice and Health Care Impacts: A

Conceptual Analysis. Working paper, Research Project Number CMD-06-02, Consumer and

Market Demand Agricultural Policy Research Network.

http://www.consumerdemand.re.ualberta.ca/Publications/CMD%2006-02.pdf

194

Appendix 1: Post Hoc Estimation for the CL Model with Interaction Effects between the

Main Attributes

Variable Coefficient t-ratio

Price -1.468*** -46.887

Function Claim 0.235** 2.056

Risk Reduction Claim 0.682*** 6.686

Disease Prevention Claim 0.536*** 5.347

Heart Symbol 0.134** 2.365

Government Verification 0.327*** 5.004

Third Party Verification 0.309*** 4.770

Omega-3 0.355*** 3.727

No Purchase -6.133*** -57.976

FC*Heart 0.146 1.241

Log Likelihood Function = -5274.259; Pseudo-R2 = 0.25

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

This table is an example to show that the results in Table 5.4 are not robust, since when all

the insignificant interaction variables from Table 5.4 are removed and the model re-estimated,

the originally marginally significant coefficient (FC*Heart) becomes insignificant. The reason

might be that the coefficient of FC*Heart, reported in Table 5.4, is only borderline significant at

the 10% level.

195

Appendix 2: Estimation Results of the RPL Model Including Reference-Dependent Effects

Variable Coefficient t-ratio WTP

($/2 Litre)

t-ratio

Mean value of Random parameters in utility function PriceGain 0.14 1.23 - - PriceLoss -0.71*** -6.27 - -

Omega3Gain -2.79*** -3.49 -1.05*** -3.50 Omega3Loss -2.38** -2.19 -0.90** -2.19

Mean value of Fixed parameters in utility function Price -2.65*** -29.88 - -

Function Claim 0.38** 2.51 0.14** 2.50 Risk Reduction Claim 1.29*** 8.30 0.49*** 8.23

Disease Prevention Claim 1.06*** 7.00 0.40*** 6.95 Heart Symbol 0.34*** 5.47 0.13*** 5.42

Government Verification 0.94*** 9.17 0.36*** 8.95 Third Party 0.74*** 7.00 0.28*** 7.00

Omega-3 4.88*** 5.53 1.84*** 5.43 No Purchase -9.71*** -36.23 -3.67*** -88.88

Standard deviations of random parameters Sd-PriceGain 0.89*** 10.88 - - Sd-PriceLoss 1.12*** 13.59 - -

Sd-Omega3Gain 4.40*** 17.03 - - Sd-Omega3Loss 3.50*** 5.53 - -

Log Likelihood Function = -3485.646; Pseudo-R2 = 0.575

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

This table presents RPL model estimates including reference dependent effects. It should be

clear that these results are very similar to the first CL model in Table 6.2(a), with the exception

that the magnitude of the coefficient for the variable Omega-3Gain is slightly larger than the

coefficient of the variable Omega-3Loss. Therefore, this result seems to be less consistent with

the prediction of Prospect Theory, which assumes that the effect of a looming loss is larger than

the effect of a looming gain. However, according to the LR test, the goodness of fit for this RPL

model is better than the first CL model in Table 6.2(a). Two reasons might help to explain the

results. First, as indicated by the standard deviation of random parameters in this model,

respondents‘ responses to an Omega-3 gain are more heterogeneous than to an Omega-3 loss.

There might be a relatively large proportion of respondents who have negative views of an

196

Omega-3 gain when Omega-3 milk is not their regularly consumed milk product. Second, as

indicated in chapter 6, a relatively small proportion of respondents incur an Omega-3 loss.

Further research might help to explain this result, when more consumers have Omega-3 milk as

their reference point in the future.

197

Appendix 3: Post Hoc Estimation for the Unused/Unqualified Data Set

As mentioned in Chapter 3, some unqualified data were not used in the formal analysis

in this study. The unqualified data include those respondents who completed the survey too

quickly and who did not pass the ―trap‖ questions. Some post hoc estimation are examined by

using the unqualified data in this analysis. Although the researcher expected that the

unused/unqualified data set is problematic and may provide inconsistent and not robust

estimation results, it might be still interesting to conduct a further check. To the researcher‘s

knowledge, this type of work has been rarely done in the literature.

In the choice experiment section of the survey, two ―trap‖ questions were added to weed

out respondents who gave inconsistent or illogical answers. One ―trap‖ question was an

asymmetric dominated choice set with a logical answer. The other ―trap‖ question was an

identical repeat of an earlier choice set, which was used to check whether a respondent‘s answers

were consistent. This idea has been mentioned by Fowler (1995), who gave a set of evaluative

strategies to find out how well answers to survey questions produce valid measurements in one

of his well-known survey research methods textbooks. One of the strategies Fowler

recommended was to measure the consistency of answers of the same respondents at two points

in time. Fowler (1995) said: ―reinterviewing a sample of respondents, asking the same questions

twice and comparing the results, can provide useful information about the validity and reliability

of answers … when questions pertain to situations or events unlikely to change between the two

data collections, differences in answers to the same question can be inferred to imply error.‖

Using trap questions to flag the problematic respondents was also highly recommended by the

marketing research company who administrated the online survey. Therefore, two ―trap‖

questions were used in this study as indicators to enhance data quality. In total, there were 95

198

(10%) respondents who provided inconsistent or illogical answers to the two ―trap‖ questions.

Their data were considered invalid and were removed from the data analysis.

The respondents who completed the survey too quickly are also weeded out, since they

may not thoughtfully consider the questions. They may perceive an incentive to complete the

questionnaire as quickly as possible and collect the benefits (e.g. sweepstakes). Their approach

might be simply to provide the appearance of compliance to minimize effort in responding to

surveys (Krosnick and Alwin, 1987 and Krosnick, 1991). Some researchers, such as Malhotra

(2008), found that it is problematic to have respondents complete surveys too quickly in self-

administrated web surveys since they may not have considered the questions thoughtfully. More

specifically, Malhotra (2008) found that low-education respondents who completed the survey

most quickly were most prone to primacy effects (bias toward selecting earlier response choices)

when completing items with unipolar rating scales. He suggested that extremely quick

completion time may be a valuable criterion in filtering out participants or their data. He also

suggested that the reported results of statistical models should be robust to the removal of

completion time outliers. Galesic and Basnjak (2009) supported the above statements by finding

that answers to questions positioned later in surveys were completed faster, shorter and more

uniform than those positioned at the beginning.

Considering the length of the survey in this study, based on the pilot study results,

respondents can rarely complete the survey less than 15 minutes. Therefore, respondents who

completed the survey too quickly (within 10 minutes) were not included in the sample. In total,

there were 5% (47 respondents) were found to answer the survey too quickly, which is consistent

with Malhotra‘s study in 2008.

199

The followings are some post hoc estimation by using the unqualified data of those

respondents who completed the survey too quickly and who did not pass the ―trap‖ questions.

Table Appendix 4(1): Base Model: CL Estimations and WTP

Using the Unqualified Data Sample (n = 142)

Variable Coefficient t-ratio WTP

($/2 Litres)

t-ratio

Price -0.842*** -15.641 - -

Function Claim 0.056 0.274 0.066 0.274

Risk Reduction Claim 0.416** 2.066 0.494** 2.058

Disease Prevention

Claim

0.706*** 3.543 0.838*** 3.489

Heart Symbol 0.249*** 2.996 0.296*** 2.970

Government

Verification

0.465*** 3.691 0.552*** 3.632

Third Party

Verification

0.465*** 3.568 0.552*** 3.537

Omega-3 0.195 1.053 0.231 1.067

No Purchase -4.171*** -20.019 -4.954*** -19.876

Log Likelihood Function = -1204.867; Pseudo-R2=0.12

*, ** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

This base model is estimated by using the unqualified data sample, and a similar model

has been estimated in Table 5.2 in Chapter 5 by using the qualified data sample. Compared with

the results in Table 5.2, one of the major differences is that the coefficients of variable Function

Claim and Omega-3 are not significant anymore, indicating respondents in the unused data set do

not value the function claim or Omega-3 attributes to be present in milk products. The second

important difference is that the Pseudo-R2 is only 0.12 in this model (0.25 in Table 5.2),

indicating that the goodness of fit of this model is poor.

200

Table Appendix 4(2): Base Model: RPL and WTP Estimates Using the Unqualified Data

Sample (n = 142)

Variable Coefficient t-ratio WTP

($/2 litres)

t-ratio

Mean value of Random parameters in utility function

Omega-3 1.190*** 3.272 1.046*** 3.347

Disease Prevention Claim 0.830*** 3.351 0.730*** 3.321

Government Verification 0.794*** 4.799 0.698*** 4.858

Mean value of Fixed parameters in utility function

Price -1.137*** -14.641 - -

Function Claim -0.091 -0.377 -0.080 -0.378

Risk Reduction Claim 0.352 1.512 0.309 1.505

Heart Symbol 0.324*** 3.392 0.285*** 3.414

Third Party Verification 0.561*** 3.540 0.493*** 3.565

No Purchase -5.042*** -19.060 -4.433*** -23.631

Standard deviations of random parameters

Sd-Omega-3 2.879*** 9.403 - -

Sd-Disease Prevention

Claim 0.957*** 3.970

- -

Sd-Government

Verification 0.560*** 1.879

- -

Log Likelihood Function = -1004.737; Pseudo-R2 = 0.36

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

Table Appendix 4(2) presents the results of the base model estimated by the Random

Parameter Logit (RPL) model using the unqualified data set. Compared with the results of the

CL model in Table Appendix 4(1), according to the LR test of the Log Likelihood Function and

the Pseudo-R2, the goodness of fit of this model has a significant improvement. However,

different from the results of the CL model in Table Appendix 4(1), the coefficient of Risk

Reduction Claim is not significant anymore and the coefficient of Omega-3 become significant

in this RPL model, indicating that respondents in the unused data set prefer Omega-3 attribute

and do not value the risk reduction claim to be present on milk products. The results of the two

models using the same data set are not consistent. These inconsistent results indicate that the

201

estimation results are not robust for the unused/unqualified data set. As expected, the quality of

the unused data set is low and the estimation are not reliable.

Table Appendix 4(3): Socio-Demographic Characteristics

of the Participants in the Qualified and Unqualified Data Sample

Demographic Characteristics Qualified Sample

(n = 740)

Unqualified Sample

(n = 142)

Age 49.3 43.6

Gender (Female) 67.8% 62%

Size of Household 2.49 2.71

Income (mean) $58,953 $61,082

Education

High school or less 35.3% 38%

College certificate or trade diploma 40.7% 38.7%

University Bachelors degree 19.1% 16.2%

University Masters degree or higher 5% 7%

Table Appendix 4(3) shows a comparison of the socio-demographic characteristics of the

respondents in the qualified and unqualified data sample. Overall, no significantly statistical

differences have been found in these two data samples. However, the differences for

characteristics of Age and Gender are borderline significant. The average age is 5.7 years

younger and there are 5% more male participants in the unqualified sample. No statistical

differences have been found for the characteristics of Income, Education and Size of Household

in these two samples.

When taking a close look at the specific answers to survey questions responded by those

participants in the unqualified data sample. Quite a few respondents gave the answers that

followed certain ‗fixed patterns‘, such as ―C,C,…,C,C‖ and ―A,B,A,B,…,A,B‖. This observation

might help to explain that some of the respondents may not thoughtfully consider the questions.

Therefore, considering all of above findings, it is necessary to weed out the ―noise‖ data of those

respondents who completed the survey too quickly and who did not pass the ―trap‖ questions.

202

Someone might ask: ―how will the estimation look like if combining both the qualified

and unqualified data together?‖ The following estimations are derived from the combined/total

data sample.

Table Appendix 4(4): Base Model: CL Estimations and WTP Using the Total Data Sample

(n = 882)

Variable Coefficient t-ratio WTP

($/2 Litres)

t-ratio

Price -1.330*** -50.492 - -

Function Claim 0.239*** 2.621 0.179*** 2.619

Risk Reduction Claim 0.609*** 6.710 0.458*** 6.740

Disease Prevention

Claim 0.554*** 6.209 0.417*** 6.204

Heart Symbol 0.193*** 4.872 0.145*** 4.864

Government

Verification 0.372*** 6.493 0.280*** 6.472

Third Party

Verification 0.346*** 6.013 0.260*** 6.020

Omega-3 0.285*** 3.503 0.215*** 3.546

No Purchase -5.742*** -61.811 -4.317*** -68.130

Log Likelihood Function =-6539.238 ; Pseudo-R2=0.22

*, ** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

Table Appendix 4(4) presents the base CL model which is estimated by using the total data

sample. Compared with the results in a similar CL model in Table 5.2 using the qualified data

sample, the results in these two models are generally consistent, although the goodness of fit of

the model in Table 5.2 is better. However, the results are not consistent with the following RPL

estimate.

203

Table Appendix 4(5): Base Model: RPL and WTP Estimates Using the Total Data Sample

(n = 882)

Variable Coefficient t-ratio WTP

($/2 litres)

t-ratio

Mean value of Random parameters in utility function

Omega-3 1.222*** 6.983 0.561*** 7.107

Function Claim 0.116 0.881 0.053 0.880

Disease Prevention Claim 0.924*** 7.241 0.424*** 7.290

Heart Symbol 0.300*** 5.520 0.138*** 5.547

Government Verification 0.944*** 9.981 0.434*** 10.219

Mean value of Fixed parameters in utility function

Price -2.177*** -40.596 - -

Risk Reduction Claim 0.802*** 6.509 0.368*** 6.560

Third Party Verification 0.592*** 7.087 0.272*** 7.183

No Purchase -8.297*** -49.291 -3.811*** -93.133

Standard deviations of random parameters

Sd-Omega-3 4.042*** 22.665 - -

Sd-Function Claim 1.030*** 8.638 - -

Sd-Disease Prevention

Claim 1.064*** 8.729

- -

Sd-Heart Symbol 0.312** 2.303 - -

Sd-Government

Verification 1.172*** 10.170

- -

Log Likelihood Function =-4915.367; Pseudo-R2 = 0.50

*,** and *** indicate significant at the 10%, 5% and 1% levels, respectively.

Table Appendix 4(5) shows the RPL model using the total data sample. Compared with

the results in the RPL model in Table 5.6 using the qualified data sample, the major difference is

that the coefficient of variable Function Claim is not significant anymore, indicating in general,

the respondents in the total sample do not value the function claim on milk products. In addition,

the variable Function Claim has fixed parameters in Table 5.6, but has random parameter with

significant standard deviation in table Appendix 4(5), indicating that respondents in the total

sample have strong variation in their preference for whether the function claim is present on milk

products. Furthermore, compared the results in Table Appendix 4(4), the coefficient of variable

204

Function Claim is not significant anymore in this model. The results of the two models using the

same data set are not consistent, indicating that the results are not robust by using the total data

sample, caused by some ―noise‖ data in the unqualified data set. When using the qualified data

sample, the estimation results are robust and consistent as shown in Chapter 5. Therefore, it is

reasonable for this study to remove those ―noise‖ data from the total data set and use the

qualified data sample in the formal analysis.

The above findings should have some important implications to improve data quality and

produce valid measurements in terms of web survey methodology.

205

Appendix 4: The Survey

Dear respondents,

This survey is being conducted by researchers in the Department of Bioresource Policy Business & Economics

at the University of Saskatchewan. The purpose of the survey is to examine peoples‘ food choices including

attitudes toward diet and health.

The information collected in this survey will help us to better understand the factors which affect consumers‘

food consumption and purchase decisions, and how consumers make decisions about foods with different health

effects.

Your responses and opinions are very important to us and they will be completely anonymous and confidential.

All data will be aggregated and individual respondents will not be identifiable. The information from the survey will

be used only for the purposes of academic research. If you have any other questions, please do not hesitate to contact

us. Thank you.

Sincerely,

Ningning Zou (Researcher)

Ph.D. Candidate in Department of Bioresource Policy Business and Economics

University of Saskatchewan

Address: 51 Campus Dr. Saskatoon, S7N 5A8

Tel: (306) 880-1568, E-mail: niz206@mail.usask.ca

Dr. Jill E. Hobbs (Supervisor)

Professor and Department Head,

Department of Bioresource Policy, Business & Economics,

University of Saskatchewan

Address: 51 Campus Dr. Saskatoon, S7N 5A8

Tel: (306) 966-2445 Fax: (306) 966-8413, E-mail: jill.hobbs@usask.ca

206

[Screener Question 1]

Do you consume (cows) milk at least once per month?

A. Yes

B. No

[Screener Question 2]

Are you one of the primary food grocery shoppers in your household?

A. Yes

B. No

Respondents that answer YES to the above 2 questions should proceed with the

survey

Those answering „no‟ should exit from the survey

207

Q1. What size of milk do you typically purchase at the grocery store? (Please check all that

apply)

A. Less than 1 litre carton of milk

B. 1 litre carton of milk

C. 2 litre carton/jug of milk

D. 4 litre jug of milk

E. Other, please specify_______

Q2. Approximately how much do you usually pay for the milk you typically purchase at the

grocery store? (Please complete all that apply)

A. Less than 1 litre carton of milk $_______

B. 1 litre carton of milk $_______

C. 2 litre carton/jug of milk $_______

D. 4 litre jug of milk $_______

E. Other, please specify_______ $_______

Q3. Approximately how much milk does your household usually purchase in a typical week?

______________ Litres

208

Q4. What is the fat level of the milk you typically purchase? (Please check all that apply)

A. 0% (Skim)

B. 1%

C. 2%

D. 3.25% (Homo)

Q5. What is the price of the 2-litre carton of milk you typically purchase?

(If you don‟t often purchase a 2-litre carton of milk, how much would you expect to pay for a 2-

litre carton of milk?)

$____________

(OR pick from a range below if you are unsure)

A. $1.99 or less

B. $2.00 - $2.99

C. $3.00 - $3.99

D. $4.00 - $4.99

E. $5.00 or more

209

Q6. How important are the following when you purchase milk?

Not at all

Important

1

2 3 4 5 6

Very

Important

7

A. Price

B. Additional health

Ingredients (e.g. milk

enriched with omega-3)

C. Fat level

D. Health Claims

E. Brand

F. Nutrition

G. Food Safety

210

Choice Experiment

In the next few questions we would like you to imagine that you are in a grocery store

buying a 2-litre carton of milk. We are going to show you examples of different types of 2

litre cartons of milk with the following features:

Milk Product Features

Feature Explanation

1. Additional

Ingredient

( )

“Omega3” on the package means the milk product was

enriched with Omega-3.

2.Health Claims Some of the milk products feature a health claim.

3.Verifying

Organization

of Health Claims

For some of the milk products, an organization has verified that

the health claim is accurate. This might be Health Canada

(government) OR the Heart and Stroke Foundation (a Third

Party organization).

4. Symbol

on package

Some of the milk products feature a symbol on the package such

as a red heart.

5. Price Retail price of a 2-litre carton of milk.

211

Here is an example

Attributes Option A Option B Option C Option D

Symbol on

Package

I would not

purchase

any of these

milk

products.

Additional

Ingredient

Health

Claims

Verifying

Organizati

on of

Health

Claims

Price $3.59 $4.49 $2.69

Choices □

□ □

Option A is a 2-litre carton milk enriched with Omega3. It has a health claim “Reduces the

Risks of Heart Disease and Cancer”, verified by the Heart & Stroke Foundation. This carton

of milk costs $3.59.

Option B is a 2-litre carton milk enriched with Omega3. It has a health claim “Helps to

Prevent Coronary Heart Disease and Cancer”, verified by Health Canada. This carton of

milk costs $4.49.

Option C is a carton of 2-litre regular milk and costs $2.69.

Option D can be selected if you would not buy options A, B, or C.

In this example, option B was chosen.

212

For the following eight questions, assume that you are shopping in a real store and

plan to buy a milk product. You will be asked to choose one option each time. Please do not

compare the options on different screens.

Q7. Assume that the options below are the only milk products available in the grocery store

on that shopping trip and that this is the fat level and brand of milk that you typically

purchase. Please choose ONE option from the following four choices.

Attribute

s

Option A Option B Option C Option D

Symbol

I would

not

purchase

any of

these

milk

products.

Ingredien

t

Health

Claims

Verifying

Organiza

tion of

Health

Claims

Price $4.49 $2.69 $1.99

Choices □ □ □ □

213

Q15. How much would you expect to pay for a 2-litre carton of milk enriched with Omega-3?

$________

(OR – choose from the range below)

A. $1.99 or less

B. $2.00 - $2.99

C. $3.00 - $3.99

D. $4.00 - $4.99

E. $5.00 or more

Q16. How much would you expect to pay for a 2-litre carton milk enriched with Omega-3?

A. When the retail price of a 2 litre carton of

regular of milk is $2.69 $____________________

B. When the retail price of a 2 litre carton of

regular of milk is $1.99 $____________________

C. When the retail price of a 2 litre carton of

regular of milk is $3.59 $____________________

214

Q17. Please tell us about the milk you typically purchase:

A. Does the milk you typically purchase have a health

claim on its package? Yes No I don‟t know

B. Does the milk you typically purchase have a general

health claim on its package? (e.g. „Good for your

heart‟)

Yes No I don‟t know

C. Does the milk you typically purchase have a risk

reduction claim on its package? (e.g. „Reduces the risk

of heart disease‟)

Yes No I don‟t know

D. Does the milk you typically purchase have a symbol

implying a health benefit (e.g. such as a „red heart‟) on

the package?

Yes No I don‟t know

E. Does the milk you typically purchase have the Heart

& Stroke Foundation „Health check‟ symbol on its

package?

Yes No I don‟t know

F. Is the milk you typically purchase Omega-3-

enriched milk? Yes No

I don‟t

Know

G. Is the milk you typically purchase enriched with a

health ingredient other than Omega-3, such as calcium

or probiotics, etc? (please specify________)

Yes No I don‟t

Know

215

Q18. What is the primary dietary source from which you obtain Omega-3? (Please check one

only)

A. Omega-3 Nutritional Supplement.

B. Milk enriched with Omega-3.

C. Other foods containing Omega 3, but not milk.

D. I rarely consume anything with Omega3.

E. I never consume anything with Omega3.

F. I don‟t know.

The following questions ask about food labels and how you find out information about the

food you consume

Q19. Below is a list of statements, please indicate how much you agree or disagree on the scale provided

Strongly

Disagree

1

2

3

4

5

6

Strongly

Agree

7

A. I regularly read labels on food I

purchase.

B. I trust nutrition labels on food

products.

C. The health claims on food

products are accurate.

D. I trust new food products.

E. I pay attention to symbols on food

packaging (e.g. a red heart) rather

than health claims.

F. I am a very trusting person.

216

Q20. Where do you typically get health information about food products? (Please check all that apply.)

A. Professionals, such as doctors, nutritionists, etc. □

B. Health organization such as Heart and Stroke Foundation of

Canada, etc. □

C. Government institution, such as Health Canada, etc. □

D. Media, such as TV, newspapers, radio, magazines □

E. Books □

F. Food labels □

G. Food companies/ industry □

H. Grocery stores □

I. Friends □

J. Internet □

K. Consumer organizations □

L. Other, please specify ___________________ □

RANKING

Please rank the top 3 sources you use most frequently.

1.___________________

2. ___________________

3.____________________

217

Q21. How much do you trust the following sources for accurate health information?

Don‟t trust at

all

1

2

3

4

5

6

Completely

Trust

7

A. Professionals, such as

doctors, nutritionists, etc.

B. Health organization such

as Heart and Stroke

Foundation of Canada,

etc.

C. Government institution,

such as Health Canada,

etc.

D. Media, such as TV,

newspapers, radio,

magazines

E. Books

F. Food labels

G. Food companies/ industry

H. Grocery stores

I. Friends

J. Internet

K. Consumer organizations

Q22. Have you ever heard of the term „functional food‟?

A. Yes

B. No

C. I don‟t know.

218

Definition of Functional Foods:

Health Canada (1999) defines that functional foods are similar in appearance to, or may be,

a conventional food, consumed as part of a usual diet, and are demonstrated to have

physiological benefits and/or reduce the risk of chronic disease beyond basic nutritional

functions.

Q23. These are some examples of functional foods. Please indicate how often you consume these products.

Never

Consume

Tried Once or

Twice

Occasionall

y Consume

Regularly

Consume

A. Omega-3 Milk

B. Milk enriched with Calcium

C. Omega-3 Eggs

D. Omega-3 Cheese

E. Omega-3 Yogurt

F. Probiotic Yogurt

G. Juice enriched with Vitamins

( e.g. Vitamin C)

H. Cereals enriched with Minerals

(e.g. Calcium)

219

Q24. Please indicate how much you agree or disagree with the following statements?

Strongly

Disagree 1

2

3

4

5

6

Strongly

Agree

7

A. Functional foods can maintain overall

wellbeing and improve long-term health.

B. Functional foods may reduce the risk of

certain chronic diseases.

C. Functional foods may prevent certain

diseases.

D. Functional foods are necessary for a

healthy diet and should be consumed

regularly.

E. I am concerned about whether

functional foods are safe.

F. I am knowledgeable about health and

nutrition.

G. My friends/ relatives ask me for health

or nutrition advice.

H. Cardiovascular disease and cancer are

the two leading causes of death in Canada.

I. Probiotic ingredients help digestive

health.

220

Q25. Please indicate the extent to which you agree or disagree with the following

statements.

Strongly

Disagree

1

2

3

4

5

6

Strongly

Agree

7

A. I often introduce foods to my diet

which may provide health benefits.

B. I often eat foods with added

vitamins and supplements.

C. Eating is a better way to obtain

health benefits than taking nutritional

supplement like vitamins.

D. Milk is a good product and I try to

include it in my diet.

E. I often eat a lot of fruit and

vegetables.

F. I often eat foods containing fibre.

G. I often eat fast food.

H. I often consume too much fat in

my meals.

I. I often consume too much sugar in

my meals.

221

Q26. What kind of role do you believe that your diet (i.e. what you eat) plays in your overall health?

Not important at all

1

2

3

4

5

6

Vitally Important

7

□ □ □ □ □ □ □

Q27. How would you describe your current health status?

Poor

1 2 3 4 5 6

Excellent

7

□ □ □ □ □ □ □

Q28. How much control do you believe that you have over your own health?

No Control at all

1 2 3 4 5 6

Complete Control

7

□ □ □ □ □ □ □

222

Q29. Please indicate whether you suffer from or have a family history of the following health

problem(s).

yourself your family members

A. Cancer □ □

B. Cardiovascular (heart) disease □ □

C. Diabetes □ □

D. High blood pressure □ □

E. High cholesterol □ □

F. Digestive problems □ □

G. Immune system deficiency □ □

H. Osteoporosis □ □

I. Weight control □ □

J. Other(Please specify) _________ __________

Q30. Please indicate how much you agree or disagree on the scale provided

Strongly

Disagree

1

2

3

4

5

6

Strongly

Agree

7

A. I try to prevent health

problems before I feel any

symptoms.

B. I am concerned that my family

history‟s health problem(s) might

happen to me in the future.

C. I am concerned that I am at

risk of developing heart disease or

cancer.

D. I am on a special diet because of

my doctor‟s advice.

223

Q31. In general, how much pressure/stress are you normally under?

No Stress

0 1 2 3 4 5

Lots of Stress

6

□ □ □ □ □ □ □

Q32. How often do you exercise?

A. Frequently ( 4 times per week or more)

B. Often (2 to 3 times per week)

C. Sometimes (once per week)

D. Rarely

E. Never

Q33. In general, do you think you are:

Underweight

1

2

3

About Right

4

5

6

Overweight

7

□ □ □ □ □ □ □

224

Q34. Do you smoke?

A. Yes

B. No

Q35. Are you a male or female?

A. Male

B. Female

Q36. In which year were you born? 19 ___ ___

Q37. Please indicate the number of people in your household.

0 1 2 3 4 5 or more

A. How many people live with you in your

household (Including yourself)?

B. How many seniors (aged 65 or over) live in

your household?

C. How many children less than 3 years of age

live in your household?

D. How many children older than 3 years of

age live in your household?

225

Q38. Please indicate the first three digits of your postal code:

____ ____ ____

Or please indicate in which province you are currently living.

_______________

Q39. Please indicate the highest level of education you have completed.

A High school or less

B College certificate or trade diploma

C. University Bachelors degree

D. University Masters degree or higher

Q40. For comparison purposes only, which of the following describes your annual household income

before taxes?

A. Less than $25,000

B $25,000 to $49,999

C. $50,000 to $74,999

D. $75,000 to $99,999

E. $100,000 to $124,999

F. $125,000 to $149,999

G. $150,000 to $174,999

H. $175,000 or above

226

Comments

______________________________________________________________________________

______________________________________________________________________________

______________________________________________________________________________

______________________________________________________________________________

______________________________________________________________________________

______________________________________________________________________________

______________________________________________________________________________

If you have any questions about this research or wish to be informed about the outcome of

the research please contact

Ningning Zou (PhD Candidate)

E-mail: niz206@mail.usask.ca

Dr. Jill E. Hobbs (Supervisor)

Email: jill.hobbs@usask.ca

Department of Bioresource Policy, Business & Economics

University of Saskatchewan

51 Campus Dr.

Saskatoon, SK

S7N 5A8

Ph: (306) 966-2445

Fax: (306) 966-8413